JP2021160714A - Vehicle control device and vehicle control method - Google Patents

Abstract

To provide a vehicle control device and a vehicle control method capable of achieving both of a safety ratio and comfort.SOLUTION: A vehicle control device 100 includes: an action plan generation part 200 to generate an action plan of automatic driving of an own vehicle 1; and distance detection parts (a detection device DD, a vehicle sensor 60, and an automatic driving control part 120) to output detection information on detection of a target. The action plan generation part 200 includes a collision rate map setting part 250 to determine a collision rate map that is a two-dimensional map displaying distribution of collision rates showing collision possibility between the own vehicle 1 and the obstacle in a two-dimensional space of position and speed when the distance detection parts detects an obstacle. The action plan generation part 200 generates a current action plan on the basis of the collision rate map, a predetermined target collision rate, and current position and speed of the own vehicle 1.SELECTED DRAWING: Figure 10A

Description

The present invention relates to a vehicle control device and a vehicle control method.

In Patent Document 1, the speed change of the vehicle is based on the position error probability distribution and the data in which the acceleration or the gradient of the acceleration that does not give the driver a sense of discomfort according to the distance from the shift start point to the target point is registered in advance. The target speed calculation unit that calculates the target speed value to the target point for each point so that is a continuous curve, detects the speed of the vehicle, and controls the drive torque so that it becomes the target speed value to control the speed of the vehicle. A vehicle speed control device including a speed control unit for controlling is described.

Patent Document 2 describes an object detecting means for detecting an object existing in the vicinity of the own vehicle and a collision possibility determining means for determining the collision possibility between the object detected by the object detecting means and the own vehicle for each discrete time. , A collision mitigation device including a collision impact reducing means for controlling to reduce the impact of a collision based on the collision possibility determined by the collision possibility determining means is described.

Japanese Patent No. 4996400 Japanese Patent No. 4967840

The technique of Patent Document 1 controls the speed of the vehicle from the position error probability distribution. Further, the technique of Patent Document 2 is for determining the possibility of collision with the own vehicle. Both techniques control the speed of the vehicle from the position error probability distribution and the possibility of collision, and are not associated with the action plan of autonomous driving. Therefore, when the action plan is determined using the sensor, there is a problem that the recognition distance is insufficient and the recognition distance cannot be extended while maintaining high reliability of the sensor. In addition, there is a problem that the level of safety is clarified, the reliability (accuracy, detection rate) of the sensor, and the accuracy of the action plan cannot be quantified.

The present invention has been made in view of such circumstances, and an object of the present invention is to provide a vehicle control device and a vehicle control method capable of achieving both a safety factor and comfort.

In order to solve the above problems, the vehicle control device according to the present invention is a vehicle control device that controls a vehicle, and is based on an action plan generation unit that generates an action plan for automatic driving of the vehicle and the action plan. The action plan generation unit is provided with at least a vehicle behavior control unit that controls the speed of the vehicle and a distance detection unit that detects a target and outputs detection information related to the detection of the target. The maximum deceleration of the vehicle is set, and when the distance detection unit detects an obstacle, the action plan generation unit may collide with the vehicle in a two-dimensional space of position and speed. The collision rate map includes a collision rate map setting unit that determines a collision rate map, which is a two-dimensional map showing the distribution of the collision rate, and the collision rate map is a combination of a predetermined target collision rate, the maximum deceleration, and the detection information. It is created on the premise of the target stop position set based on the above, and the action plan generation unit is currently based on the collision rate map, the predetermined target collision rate, and the current position and speed of the vehicle. It is characterized by generating an action plan for.

According to the present invention, it is possible to provide a vehicle control device and a vehicle control method capable of achieving both a safety factor and comfort.

It is a figure which shows the whole structure of the vehicle which includes the vehicle control device which concerns on embodiment of this invention. It is a functional block diagram centering on the vehicle control device which concerns on the said embodiment. It is a block diagram of the vehicle control device HMI which concerns on the said embodiment. It is a block diagram of the action plan generation part of the vehicle control device which concerns on the said embodiment. It is a figure which shows the target speed value curve when decelerating to the target stop position of the vehicle control device which concerns on the said embodiment, (a) is a figure which shows the target speed value curve when decelerating at 0.6G, (b). ) Is a diagram showing a detection rate P (D | E) in which the sensor detects an obstacle in the presence of an obstacle, and (c) is an erroneous detection in which the sensor detects an obstacle in the absence of an obstacle. The rate P (D | E￣) is shown, and (d) is a diagram showing an example of a deceleration profile of an action plan in the case of automatic driving. It is a figure explaining the action plan based on the reliability of a sensor in the determination of the target speed value curve of FIG. It is a figure which shows the case of setting. It is a figure explaining the action plan based on the reliability of a sensor in the determination of the target speed value curve of FIG. 5, and the detection value of a sensor is set to be used in the region of high reliability in the determination of the action plan of FIG. It is a figure which shows the case. It is a figure explaining the direction which tries to solve the trouble in FIG. 6 and FIG. It is explanatory drawing explaining the collision rate which decelerates to the front running vehicle position (target stop position) of the vehicle control device which concerns on the said embodiment and prevents a collision, and FIG. shows a collision rate map to prevent, (b) is (a) collision ratio P of | diagram illustrating a (C _{S t)} with gradation representation, (c), the collision when the speed 0 of the (a) It is a figure which shows the image of the peak of a rate. A collision rate is set when the vehicle of the vehicle control device according to the above embodiment decelerates to the preceding vehicle position (target stop position) to prevent a collision, and a target speed value based on an action plan in which this collision rate is introduced. It is a figure which shows a curve. It is a figure which shows the fusion accuracy reliability which is determined by the sensor configuration which detected the target, and the detection time when the detection device of a vehicle detects a target. When the vehicle detection device according to the above embodiment detects a target, the command value obtained by the probability distribution based on the fusion accuracy reliability determined by the sensor configuration and the detection time that detected the target and the command value achievement probability density distribution. It is a figure which shows the change of the offset amount of. It is a command value determination flow of the action plan of the vehicle control device which concerns on the said embodiment. It is a figure which shows the relationship between the width of the probability distribution based on the fusion accuracy reliability of the vehicle control device which concerns on the said embodiment, and the offset amount. It is a flowchart of the vehicle control of the vehicle control device which concerns on the said embodiment. It is a figure which shows the collision rate map of the vehicle control device which concerns on the said embodiment, FIG. The figure which expresses the collision rate expressed by a) in a gradation expression as the reliability rate in observation, (c) is a figure which shows the image of the peak of the collision rate at a speed 0 in the collision rate map of (a). .. It is a figure explaining the relationship between the action plan of the vehicle control device which concerns on the said embodiment, and the risk of a collision. It is a figure which shows the error distribution of the sensor of the vehicle control device which concerns on the said embodiment. It is a figure explaining the influence which the action plan of the vehicle control device and the sensor which concerns on the said embodiment have on a collision risk, and (a) is the collision rate when the maximum braking-0.6G and the error standard deviation of a sensor are σ1. The figure showing the map, (b) is the figure showing the target collision rate when the action plan setting of the collision rate map of (a) is changed, and (c) is the sensor performance of the collision rate map of (a) deteriorated. It is a figure which shows the target collision rate at the time of this. It is a figure explaining the reliability (detection rate) utilization method of the plurality of sensors of the vehicle control device which concerns on the said embodiment, (a) is a figure which shows the example of the case where a plurality of sensors are detected by AND, (b) is a figure which shows the example. , (C) is a diagram showing an example in the case of detecting a plurality of sensors by OR, and (c) is a diagram showing an example in the case of performing braking according to the detection state of the plurality of sensors. It is a state transition diagram explaining the change of the collision rate and the discomfort rate by the algorithm α of the vehicle control device which concerns on the said embodiment. It is a figure which shows the specific example of the probability of the state transition caused by the action shown in FIG. It is a figure explaining the state transition tree with respect to the action of the vehicle control device which concerns on the said embodiment. It is a figure explaining the change of the detection rate on the state transition tree with respect to the action of FIG. 21, and FIG. 21 (a) shows the example of transition from the first state to the state transition tree of FIG. FIG. 21 (b) is a diagram showing an example in which the speed is slightly reduced from the initial state in the state transition tree of FIG. 21, and FIG. 21 (c) is a diagram showing the speed from the initial state in the state transition tree of FIG. It is a figure which shows the example which dropped and transitioned. It is a figure explaining the relationship between the state transition tree and the algorithm with respect to the action of FIG. 21. It is a figure explaining the continuous handling of the collision rate by the algorithm α of the vehicle control device which concerns on the said embodiment. It is an image diagram which shows the outline of the action plan of the vehicle control device which concerns on the said embodiment. FIG. 25 is an image diagram showing the results of the action plan of FIG. 25 and the cruising operation executed when no obstacle is detected, represented by a matrix consisting of obstacle detection and no obstacle detection, and presence of obstacles and no obstacles. It is an image diagram explaining AND detection in the action plan using the redundancy of two sensors of the vehicle control device which concerns on the said embodiment. It is an image diagram explaining OR detection in the action plan using the redundancy of two sensors of the vehicle control device which concerns on the said embodiment. It is an image diagram explaining HALF AND detection in the action plan using the redundancy of two sensors of the vehicle control device which concerns on the said embodiment. It is a state transition diagram of the event that the collision of the vehicle control device which concerns on the said embodiment occurs. It is a figure explaining the method of obtaining the collision rate of the vehicle control device which concerns on the said embodiment, (a) is the figure which shows the target speed value curve based on the action plan which introduced the collision rate, (b) is the speed. The figure which shows the collision rate of 0, (c) is the figure which shows the error distribution of a distance detection part, and (d) is the figure which shows the detection rate of a distance detection part. FIG. 31 is a state transition diagram showing the process of collision generation in FIG. 31. It is a figure which shows the grid which expresses the collision rate map of the vehicle control device which concerns on the said Embodiment on the two-dimensional space of position and speed. It is a figure explaining that the condition at the edge of the grid of the collision rate map of the vehicle control device which concerns on the said embodiment is given. It is a figure explaining that the collision rate is obtained in order from the edge of the grid of the collision rate map of the vehicle control device which concerns on the said embodiment. It is a figure explaining the method of obtaining the collision rate by approximation from the next state of the vehicle control device which concerns on the said embodiment. It is a figure explaining the method of obtaining the collision rate by approximation from the next state of the vehicle control device which concerns on the said embodiment. It is a figure explaining the method of obtaining the collision rate by approximation from the next state of the vehicle control device which concerns on the said embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
(Embodiment)
FIG. 1 is a diagram showing an overall configuration of a vehicle including the vehicle control device 100 according to the embodiment of the present invention. The vehicle (hereinafter referred to as own vehicle 1) 1 on which the vehicle control device 100 of the present embodiment is mounted is, for example, a vehicle such as a two-wheeled vehicle, a three-wheeled vehicle, or a four-wheeled vehicle, and powers an internal combustion engine such as a diesel engine or a gasoline engine. It includes automobiles that are the source, electric vehicles that are powered by electric motors, and hybrid automobiles that have both an internal combustion engine and an electric motor. Electric vehicles are driven using, for example, power discharged by batteries such as secondary batteries, hydrogen fuel cells, metal fuel cells, alcohol fuel cells, and the like.

[Own vehicle 1]
As shown in FIG. 1, the own vehicle 1 is equipped with sensors (distance detection unit) such as a finder 20, a radar 30, a camera 40, a navigation device 50, and a vehicle control device 100.
The finder 20 is, for example, a LIDAR (Light Detection and Rangeing) that measures scattered light with respect to irradiation light and measures the distance to an object. For example, two finder 20s are installed at positions separated from each other on the left and right sides of the front, and three finder 20s are installed at the rear (a total of five finder 20s including the front and the rear).

For example, three radars 30 are installed at the front and two radars 30 are installed at the rear (a total of five radars 30 including the front and the rear). The radar 30 detects an object by, for example, an FM-CW (Frequency Modulated Continuous Wave) method.

The camera 40 is, for example, a digital camera using an individual image sensor such as a CCD (Charge Coupled Device) or a CMOS (Complementary Metal Oxide Semiconductor). The camera 40 is attached to the upper part of the front windshield, the back surface of the rearview mirror, and the like. The camera 40 periodically and repeatedly images the front of the own vehicle 1, for example. In this example, two monocular cameras are arranged side by side. The camera 40 may be a stereo camera.
The configuration shown in FIG. 1 is merely an example, and a part of the configuration may be omitted or another configuration may be added.

[Vehicle control device 100]
FIG. 2 is a functional configuration diagram centered on the vehicle control device 100 according to the present embodiment.
As shown in FIG. 2, the own vehicle 1 includes a detection device DD (sensor) (distance detection unit) including a finder 20, a radar 30, a camera 40, and the like, a navigation device 50, a communication device 55, and a vehicle sensor. A 60 (sensor) (distance detection unit), an HMI (Human Machine Interface) 70, a vehicle control device 100, a traveling driving force output device 300, a steering device 310, and a braking device 320 are mounted. These devices and devices are connected to each other by a multiplex communication line such as a CAN (Control Area Network) communication line, a serial communication line, a wireless communication network, or the like. The vehicle control device in the claims does not only refer to the "vehicle control device 100", but may include a configuration other than the vehicle control device 100 (detection device DD, HMI 70, etc.).

<Navigation device 50>
The navigation device 50 includes a GNSS (Global Navigation Satellite System) receiver, map information (navigation map), a touch panel display device that functions as a user interface, a speaker, a microphone, and the like. The navigation device 50 identifies the position of the own vehicle 1 by the GNSS receiver, and derives a route from that position to the destination specified by the user. The route derived by the navigation device 50 is provided to the target lane determination unit 110 (described later) of the vehicle control device 100. The position of the own vehicle 1 may be specified or complemented by an INS (Inertial Navigation System) using the output of the vehicle sensor 60. Further, when the vehicle control device 100 is executing the manual driving mode, the navigation device 50 guides the route to the destination by voice or navigation display.

The configuration for specifying the position of the own vehicle 1 may be provided independently of the navigation device 50. Further, the navigation device 50 may be realized by, for example, the function of a terminal device such as a smartphone or a tablet terminal owned by the user. In this case, information is transmitted and received between the terminal device and the vehicle control device 100 by wireless or wired communication.

<Communication device 55>
The communication device 55 performs wireless communication using, for example, a cellular network, a Wi-Fi network, Bluetooth (registered trademark), DSRC (Dedicated Short Range Communication), or the like. The communication device 55 wirelessly communicates with an information providing server of a system for monitoring the traffic condition of the road such as VICS (registered trademark) (Vehicle Information and Communication System), and performs wireless communication with the road on which the own vehicle 1 is traveling. Acquires information (hereinafter referred to as traffic information) indicating the traffic condition of the road to be driven. Traffic information includes traffic jam information ahead, time required at traffic jam points, accident / breakdown vehicle / construction information, speed regulation / lane regulation information, parking lot location, parking lot / service area / parking area full / empty information, etc. Information is included. Further, the communication device 55 acquires the above traffic information by communicating with a wireless beacon provided on a side band of a road or the like, or by performing vehicle-to-vehicle communication with another vehicle traveling around the own vehicle 1. It's okay. The communication device 55 is an example of a “communication unit” that acquires traffic congestion information.

<Vehicle sensor 60>
The vehicle sensor 60 includes a vehicle speed sensor that detects the vehicle speed, an acceleration sensor that detects the acceleration, a yaw rate sensor that detects the angular velocity around the vertical axis, an orientation sensor that detects the direction of the own vehicle 1, and the like. The vehicle sensor 60 may be referred to as a “sensor” in the design method / calculation formula of the action plan of the present specification.

<HMI70>
FIG. 3 is a block diagram of the HMI 70.
As shown in FIG. 3, the HMI 70 includes a configuration of a driving operation system and a configuration of a non-driving operation system. These boundaries are not clear, and the configuration of the driving operation system may include the functions of the non-driving operation system (or vice versa). The navigation device 50 and the HMI 70 described above are examples of the “output unit”.
The HMI 70 has an accelerator pedal 71, an accelerator opening sensor 72, an accelerator pedal reaction force output device 73, a brake pedal 74, a brake depression sensor (or a master pressure sensor, etc.) 75, and a shift lever 76 as a configuration of a driving operation system. The shift position sensor 77, the steering wheel 78, the steering steering angle sensor 79, the steering torque sensor 80, and other driving operation devices 81 are included.

The accelerator pedal 71 is an operator for receiving an acceleration instruction (or a deceleration instruction by a return operation) by a vehicle occupant. The accelerator opening sensor 72 detects the amount of depression of the accelerator pedal 71, and outputs an accelerator opening signal indicating the amount of depression to the vehicle control device 100.
Instead of outputting to the vehicle control device 100, it may be output directly to the traveling driving force output device 300, the steering device 310, or the brake device 320. The same applies to the configurations of other operation operation systems described below. The accelerator pedal reaction force output device 73 outputs a force (operation reaction force) in the direction opposite to the operation direction to the accelerator pedal 71 in response to an instruction from, for example, the vehicle control device 100.

The brake pedal 74 is an operator for receiving a deceleration instruction by a vehicle occupant. The brake step sensor 75 detects the stepping amount (or stepping force) of the brake pedal 74, and outputs a brake signal indicating the detection result to the vehicle control device 100.
The shift lever 76 is an operator for receiving an instruction to change the shift stage by the vehicle occupant. The shift position sensor 77 detects the shift stage instructed by the vehicle occupant and outputs a shift position signal indicating the detection result to the vehicle control device 100.
The steering wheel 78 is an operator for receiving a turning instruction by a vehicle occupant. The steering steering angle sensor 79 detects the operating angle of the steering wheel 78 and outputs a steering steering angle signal indicating the detection result to the vehicle control device 100. The steering torque sensor 80 detects the torque applied to the steering wheel 78 and outputs a steering torque signal indicating the detection result to the vehicle control device 100.
The other driving operation device 81 is, for example, a joystick, a button, a dial switch, a GUI (Graphical User Interface) switch, or the like. The driving operation device 81 receives an acceleration instruction, a deceleration instruction, a turning instruction, and the like, and outputs the instruction to the vehicle control device 100.

The HMI 70 has, as a non-driving operation system configuration, for example, a display device 82, a speaker 83, a contact operation detection device 84, a content reproduction device 85, various operation switches 86, a seat 88, a seat drive device 89, and a window glass 90. And the window drive device 91.
The display device 82 is, for example, an LCD (Liquid Crystal Display) or an organic EL (Electroluminescence) display device, which is attached to each part of the instrument panel, an arbitrary place facing the passenger seat or the rear seat, and the like. Further, the display device 82 may be a HUD (Head Up Display) that projects an image on a front windshield or other windows. The speaker 83 outputs sound. When the display device 82 is a touch panel, the contact operation detection device 84 detects the contact position (touch position) on the display screen of the display device 82 and outputs the contact position (touch position) to the vehicle control device 100. If the display device 82 is not a touch panel, the contact operation detection device 84 may be omitted.

The content playback device 85 includes, for example, a DVD (Digital Versailles Disc) playback device, a CD (Compact Disc) playback device, a television receiver, a generation device for various guide images, and the like. The display device 82, the speaker 83, the contact operation detection device 84, and the content reproduction device 85 may have a configuration in which a part or all of them are common to the navigation device 50.
The various operation switches 86 are arranged at arbitrary positions in the vehicle interior. The various operation switches 86 include an automatic operation changeover switch 87 that instructs the start (or future start) and stop of the automatic operation. The automatic operation changeover switch 87 may be either a GUI (Graphical User Interface) switch or a mechanical switch. Further, the various operation switches 86 may include a switch for driving the seat drive device 89 and the window drive device 91.

The seat 88 is a seat on which a vehicle occupant sits. The seat driving device 89 freely drives the reclining angle, the position in the front-rear direction, the yaw angle, and the like of the seat 88. The window glass 90 is provided on each door, for example. The window driving device 91 opens and closes the window glass 90.
The vehicle interior camera 95 is a digital camera that uses a solid-state image sensor such as a CCD or CMOS. The vehicle interior camera 95 is attached to a rearview mirror, a steering boss, an instrument panel, or the like at a position where at least the head of a vehicle occupant who performs a driving operation can be imaged. The vehicle interior camera 95 periodically and repeatedly images the vehicle occupant, for example.

Returning to FIG. 2, the vehicle control device 100 is implemented, for example, by one or more processors or hardware having equivalent functions. The vehicle control device 100 is configured by combining a processor such as a CPU (Central Processing Unit), a storage device, an ECU (Electronic Control Unit) to which a communication interface is connected by an internal bus, an MPU (Micro-Processing Unit), or the like. It may be.

The vehicle control device 100 includes a target lane determination unit 110, an automatic driving control unit 120 (distance detection unit), an automatic driving mode control unit 130, a recognition unit 140, a switching control unit 150, and a travel control unit 160 (vehicle). A behavior control unit), an HMI control unit 170, and a storage unit 180 are provided.
A part or all of the target lane determination unit 110, each unit of the automatic driving control unit 120, and the travel control unit 160 is realized by executing a program (software) by the processor. In addition, some or all of these may be realized by hardware such as LSI (Large Scale Integration) or ASIC (Application Specific Integrated Circuit), or may be realized by a combination of software and hardware.

After that, when the main body is described as "○○ part is", the automatic operation control unit 120 reads each program from ROM / EEPROM (Electrically Erasable Program Read-Only Memory) and loads it into RAM as necessary. Each function (described later) shall be executed. Each program may be stored in the storage unit 180 in advance, or may be incorporated into the vehicle control device 100 when necessary via another storage medium or communication medium.

<Target lane determination unit 110>
The target lane determination unit 110 is realized by, for example, an MPU (Micro Processing Unit). The target lane determination unit 110 divides the route provided by the navigation device 50 into a plurality of blocks (for example, divides the route every 100 [m] with respect to the vehicle traveling direction), and refers to the high-precision map information 181 for each block. Determine the target lane. The target lane determination unit 110 determines, for example, which lane from the left to drive. The target lane determination unit 110 determines the target lane so that the own vehicle 1 can travel on a reasonable travel route to proceed to the branch destination, for example, when there is a branch point or a merging point on the route. .. The target lane determined by the target lane determination unit 110 is stored in the storage unit 180 as the target lane information 182.

<Automatic driving control unit 120 (distance detection unit)>
The automatic operation control unit 120 includes an automatic operation mode control unit 130, a recognition unit 140, and a switching control unit 150.

<Automatic operation mode control unit 130>
The automatic driving mode control unit 130 determines the automatic driving mode based on the operation of the vehicle occupant with respect to the HMI 70, the event determined by the action plan generation unit 200, the traveling mode determined by the track generation unit 145, and the like. The mode of automatic operation is notified to the HMI control unit 170. Further, the automatic driving mode may be set to a limit according to the performance of the detection device DD (sensor) of the own vehicle 1.

In any of the automatic operation modes, it is possible to switch (override) to the manual operation mode by operating the configuration of the operation operation system in the HMI 70. The override includes, for example, when the operation of the HMI 70 by the vehicle occupant of the own vehicle 1 continues for a predetermined time or longer, a predetermined amount of operation change (for example, the accelerator opening of the accelerator pedal 71 (described later), the brake pedal 74 (described later)). ), The steering wheel 78 (described later) or more), or when the operation on the driving operation system is performed a predetermined number of times or more.

<Recognition unit 140>
The recognition unit 140 includes a vehicle position recognition unit 141, an outside world recognition unit 142, a person detection unit 143 (detection unit), an AI (Artificial Intelligence) accelerator 144 (detection unit), and an action plan. A generation unit 200 and an orbit generation unit 145 are provided.

<Own vehicle position recognition unit 141>
The vehicle position recognition unit 141 inputs from the high-precision map information 181 stored in the storage unit 180, the finder 20, the radar 30 (sensor), the camera 40 (sensor), the navigation device 50, or the vehicle sensor 60 (sensor). Based on the information to be obtained, the lane in which the own vehicle 1 is traveling (traveling lane) and the relative position of the own vehicle 1 with respect to the traveling lane are recognized.

The own vehicle position recognition unit 141 recognizes the road marking line pattern (for example, the arrangement of solid lines and broken lines) recognized from the high-precision map information 181 and the roads around the own vehicle 1 recognized from the image captured by the camera 40. By comparing with the pattern of the lane marking, the traveling lane is recognized. In this recognition, the position of the own vehicle 1 acquired from the navigation device 50 and the processing result by the INS may be added.

<External world recognition unit 142>
Returning to FIG. 2, the outside world recognition unit 142 recognizes the positions of surrounding vehicles and the states such as speed and acceleration based on the information input from the finder 20, the radar 30, the camera 40, and the like. The peripheral vehicle is, for example, a vehicle that travels around the own vehicle 1 and travels in the same direction as the own vehicle 1. The position of the peripheral vehicle may be represented by a representative point such as the center of gravity or a corner of another vehicle, or may be represented by an area represented by the outline of the other vehicle. The "state" of the peripheral vehicle may include the acceleration of the peripheral vehicle and whether or not the vehicle is changing lanes (or whether or not the vehicle is trying to change lanes), which is grasped based on the information of the various devices. Further, the outside world recognition unit 142 may recognize the positions of guardrails, utility poles, parked vehicles, pedestrians, and other objects in addition to peripheral vehicles.

<People detection unit 143>
The person detection unit 143 detects a person from the image taken by the camera 40. Specifically, the person detection unit 143 detects a specific target (person, bicycle, etc.) in a specific area by using the AI accelerator 144. The person detection unit 143 issues a person detection request to the AI accelerator 144, and the AI accelerator 144 executes an AI calculation other than the CPU and transmits the person detection result to the person detection unit 143. Since high speed is required for human detection, the AI accelerator 144 is used for human detection. However, a mode in which the AI accelerator 144 is not used may be used.

For convenience of explanation, the camera 40 and the outside world recognition unit 142 are described separately for the person detection unit 143, but any device that can detect a specific target is sufficient, and a person or the like can be detected from an image taken by the camera 40. The image processing unit to be extracted, or the internal processing of the external world recognition unit 142 may recognize and detect a person or the like from the contour of the image. In this case, the human detection unit 143 is excluded from the recognition unit 140 in FIG.
Further, as will be described later, it is also possible to further increase the recognition probability of the person detected by the person detection unit 143 by using the VICS information obtained by the communication device 55.

<AI Accelerator 144>
The AI accelerator 144 is a dedicated processor that detects a person, and uses a computing resource other than the CPU. The AI accelerator 144 is, for example, an accelerator for signal processing using FPGA (Field Programmable Gate Array), which is an image processing by a processor enhanced with a GPU (Graphics Processing Unit). Further, the AI accelerator 144 executes the calculation of AI on a dedicated hardware (for example, GPU).

<Orbit generator 145>
When executing the lane keep event, the track generation unit 145 determines and determines one of running modes such as constant speed running, following running, low speed following running, deceleration running, curve running, and obstacle avoiding running. A track candidate is generated based on the traveling mode.
Further, the track generation unit 145 evaluates the generated track candidates from the two viewpoints of, for example, planning and safety, and selects a track to be output to the travel control unit 160. From the viewpoint of planning, for example, the trajectory is highly evaluated when the followability to the already generated plan (for example, the action plan) is high and the total length of the trajectory is short. For example, when it is desired to change lanes to the right, a track such as changing lanes to the left and returning is given a low evaluation. From the viewpoint of safety, for example, at each track point, the farther the distance between the own vehicle 1 and the object (peripheral vehicle or the like) is, and the smaller the amount of change in acceleration / deceleration or steering angle is, the higher the evaluation is.

<Switching control unit 150>
Returning to FIG. 2, the switching control unit 150 switches between the automatic operation mode and the manual operation mode based on the signal input from the automatic operation changeover switch 87 (see FIG. 3) and the like. Further, the switching control unit 150 switches from the automatic operation mode to the manual operation mode based on the operation of instructing acceleration, deceleration, or steering with respect to the configuration of the operation operation system in the HMI 70. For example, the switching control unit 150 switches from the automatic operation mode to the manual operation mode when the operation amount indicated by the signal input from the configuration of the operation operation system in the HMI 70 exceeds the threshold value for the reference time or longer. override).

Further, the switching control unit 150 may return to the automatic operation mode when no operation on the configuration of the operation operation system in the HMI 70 is detected for a predetermined time after switching to the manual operation mode by overriding. .. Further, the switching control unit 150 notifies the vehicle occupants of the handover request in advance when performing the handover control for shifting from the automatic driving mode to the manual driving mode, for example, at the scheduled end point of the automatic driving. Information is output to the HMI control unit 170.

<Traveling control unit 160>
The travel control unit 160 controls the travel driving force output device 300, the steering device 310, and the brake device 320 so that the own vehicle 1 passes through the track generated by the track generation unit 145 on time.
The travel control unit 160 has a function as a vehicle behavior control unit that controls at least the speed of the vehicle based on the action plan.

<HMI control unit 170>
When the automatic operation control unit 120 notifies the HMI control unit 170 of the automatic operation mode information, the HMI control unit 170 refers to the mode-specific operation availability information 184 (see FIG. 6 below) according to the type of the automatic operation mode. Controls the HMI 70.
The HMI control unit 170 is a device (a part or all of the navigation device 50 and the HMI 70) that is permitted to be used by referring to the mode-specific operation availability information 184 based on the mode information acquired from the automatic operation control unit 120. And, it is determined that the device is not permitted to be used. Further, the HMI control unit 170 controls whether or not the non-driving operation system HMI 70 or the navigation device 50 can be operated by the vehicle occupants based on the determination result.

For example, when the operation mode executed by the vehicle control device 100 is the manual operation mode, the vehicle occupant uses the operation system of the HMI 70 (for example, the accelerator pedal 71, the brake pedal 74, the shift lever 76, the steering wheel 78, etc.) (FIG. 3) is operated. Further, when the driving mode executed by the vehicle control device 100 is the automatic driving mode, the vehicle occupant is obliged to monitor the surroundings of the own vehicle 1.

In such a case, the HMI control unit 170 is one of the non-driving operation systems of the HMI 70 in order to prevent distraction (driver distraction) due to actions other than driving by the vehicle occupant (for example, operation of the HMI 70). Control so that operations on parts or all are not accepted. At this time, the HMI control unit 170 displays the presence of the peripheral vehicle of the own vehicle 1 recognized by the outside world recognition unit 142 and the state of the peripheral vehicle thereof in order to monitor the periphery of the own vehicle 1 (FIG. 3) may be displayed as an image or the like, and the HMI 70 may be made to accept a confirmation operation according to the scene when the own vehicle 1 is traveling.

Further, when the driving mode is automatic driving, the HMI control unit 170 relaxes the regulation of driver distraction and controls to accept the operation of the vehicle occupant with respect to the non-driving operation system that has not accepted the operation. For example, the HMI control unit 170 causes the display device 82 to display an image, the speaker 83 (see FIG. 3) to output audio, and the content playback device 85 (see FIG. 3) to reproduce content from a DVD or the like. do. The content reproduced by the content reproduction device 85 may include, for example, various contents related to entertainment such as a television program, in addition to the content stored in a DVD or the like. Further, the "content playback operation" shown in FIG. 6 described above may mean such a content operation related to entertainment and entertainment.

<Memory unit 180>
Information such as high-precision map information 181, target lane information 182, action plan information 183, and mode-specific operation availability information 184 is stored in the storage unit 180. The storage unit 180 is realized by a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), a flash memory, or the like. The program executed by the processor may be stored in the storage unit 180 in advance, or may be downloaded from an external device via an in-vehicle Internet facility or the like. Further, the program may be installed in the storage unit 180 by mounting a portable storage medium in which the program is stored in a drive device (not shown).

The high-precision map information 181 is map information with higher accuracy than the navigation map included in the navigation device 50. The high-precision map information 181 includes, for example, information on the center of the lane, information on the boundary of the lane, and the like. The above boundaries include the type, color, length, road width, road shoulder width, main line width, lane width, boundary position, boundary type (guardrail, planting, curb), zebra zone, etc. of the lane mark. Is included in the high-precision map.

Further, the high-precision map information 181 may include road information, traffic regulation information, address information (address / zip code), facility information, telephone number information, and the like. Road information includes information indicating the type of road such as highways, toll roads, national roads, and prefectural roads, the number of lanes of the road, the width of each lane, the slope of the road, and the position of the road (longitudinal, latitude, height). Includes 3D coordinates), lane curve curvature, lane confluence and branch point locations, road signs, and other information. Traffic regulation information includes information that lanes are blocked due to construction work, traffic accidents, traffic jams, and the like.

[Action plan generation unit 200]
-Basic action plan The action plan generation unit 200 sets a start point of automatic driving and / or a destination of automatic driving. The starting point of the automatic driving may be the current position of the own vehicle 1 or a point where an operation for instructing the automatic driving is performed. The action plan generation unit 200 generates an action plan in the section between the starting point and the destination of the automatic driving. Not limited to this, the action plan generation unit 200 may generate an action plan for any section.

An action plan consists of, for example, a plurality of events that are executed sequentially. The events include, for example, a deceleration event for decelerating the own vehicle 1, an acceleration event for accelerating the own vehicle 1, a lane keeping event for driving the own vehicle 1 so as not to deviate from the traveling lane, and a lane change event for changing the traveling lane. , An overtaking event that causes the own vehicle 1 to overtake the vehicle in front, a branch event that causes the own vehicle 1 to change to the desired lane at the branch point, or to drive the own vehicle 1 so as not to deviate from the current driving lane, to join the main line. A merging event that accelerates or decelerates the own vehicle 1 in the merging lane and changes the traveling lane, shifts from the manual driving mode to the automatic driving mode at the start point of automatic driving, or manually changes from the automatic driving mode at the scheduled end point of automatic driving. It includes a handover event that shifts to the operation mode and the like.

The action plan generation unit 200 sets a lane change event, a branch event, or a merging event at a position where the target lane determined by the target lane determination unit 110 is switched. The information indicating the action plan generated by the action plan generation unit 200 is stored in the storage unit 180 as the action plan information 183 (described later).

[Action plan generation unit 200]
-Vehicle action plan determined based on the collision rate FIG. 4 is a block diagram of the action plan generation unit 200.
The action plan generation unit 200 obtains a collision rate indicating the possibility of collision between an obstacle and the vehicle based on the detected distance, and generates an action plan of the vehicle based on the collision rate. The action plan generation unit 200 generates an action plan for automatic driving of the vehicle.

As shown in FIG. 4, the action plan generation unit 200 stores a collision rate map 1000 (see FIG. 14) in which an action plan having a collision rate is visualized by a two-dimensional map of position and speed. Collision rate map storage unit 1010 The target collision rate setting unit 210, the command value achievement probability density distribution estimation unit 220, the command value offset amount calculation unit 230 (offset amount calculation unit), the fusion accuracy reliability estimation unit 240, and the collision rate map setting unit. 250 and.

The action plan generation unit 200 maintains the set speed of automatic operation and allows sudden braking in a region where the collision rate on the collision rate map 1000 is lower than the target collision rate and lower than a predetermined threshold. In the collision rate high region where the collision rate on the collision rate map 1000 is lower than the target collision rate and higher than the predetermined threshold, short-time braking is repeated to generate a preliminary braking action plan to avoid sudden braking. do.

<Target collision rate setting unit 210>
The target collision rate setting unit 210 sets a predetermined target collision rate.

<Command value achievement probability density distribution estimation unit 220>
The command value achievement probability density distribution estimation unit 220 achieves the command value determined by the magnitude and certainty of the difference between the command value from the travel control unit 160 (vehicle behavior control unit) and the position actually reached by the vehicle. The probability density distribution 1001 (see FIG. 10A) is estimated.

<Command value offset amount calculation unit 230>
The command value offset amount calculation unit 230 calculates a command value offset amount 1002 (see FIG. 10A) indicating the distance between the target stop position and the obstacle.

<Fusion accuracy reliability estimation unit 240>
The fusion accuracy reliability estimation unit 240 calculates a probability distribution 1003 (see FIG. 10A) based on the fusion accuracy reliability of the recognition accuracy of the detection device DD (particularly, the distance detection sensor) (distance detection unit). The probability distribution 1003 based on the fusion accuracy reliability depends on the (distance) detection accuracy of the camera, radar, or the like. Therefore, it depends on sensor redundancy, AND detection (see FIG. 27), or OR detection (see FIG. 28), which will be described later.

<Collision rate map setting unit 250>
The collision rate map setting unit 250 determines the collision rate map according to the following equation (1).

[Traveling driving force output device 300, steering device 310, and brake device 320]
Returning to FIG. 2, the vehicle control device 100 controls the traveling driving force output device 300, the steering device 310, and the brake device 320.
<Traveling driving force output device 300>
The traveling driving force output device 300 outputs a traveling driving force (torque) for traveling the vehicle to the drive wheels. The traveling driving force output device 300 includes, for example, an engine, a transmission, and an engine ECU (Electronic Control Unit) that controls the engine when the own vehicle 1 is a vehicle powered by an internal combustion engine. In the case of an electric vehicle powered by an electric motor, it is provided with a traveling motor and a motor ECU for controlling the traveling motor. It is equipped with a motor ECU.

When the traveling driving force output device 300 includes only the engine, the engine ECU adjusts the throttle opening degree, the shift stage, and the like of the engine according to the information input from the traveling control unit 160 described later. When the traveling driving force output device 300 includes only the traveling motor, the motor ECU adjusts the duty ratio of the PWM signal given to the traveling motor according to the information input from the traveling control unit 160. When the traveling driving force output device 300 includes an engine and a traveling motor, the engine ECU and the motor ECU cooperate with each other to control the traveling driving force according to the information input from the traveling control unit 160.

<Steering device 310>
The steering device 310 includes, for example, a steering ECU and an electric motor.
The electric motor, for example, applies a force to the rack and pinion mechanism to change the direction of the steering wheel. The steering ECU drives the electric motor according to the information input from the vehicle control device 100, or the information of the steering steering angle or steering torque input, and changes the direction of the steering wheels.

<Brake device 320>
The brake device 320 is, for example, an electric servobrake device including a brake caliper, a cylinder for transmitting flood control to the brake caliper, an electric motor for generating flood control in the cylinder, and a braking control unit. The braking control unit of the electric servo brake device controls the electric motor according to the information input from the traveling control unit 160 so that the brake torque corresponding to the braking operation is output to each wheel. The electric servo brake device may be provided with a mechanism for transmitting the hydraulic pressure generated by the operation of the brake pedal to the cylinder via the master cylinder as a backup.

The brake device 320 is not limited to the electric servo brake device described above, and may be an electronically controlled hydraulic brake device. The electronically controlled hydraulic brake device controls the actuator according to the information input from the traveling control unit 160 to transmit the oil pressure of the master cylinder to the cylinder. Further, the braking device 320 may include a regenerative brake by a traveling motor that can be included in the traveling driving force output device 300.

<Distance detector>
In the present embodiment, the automatic driving control unit 120 constitutes a distance detection unit together with the detection device DD and the vehicle sensor 60. The distance detection unit outputs detection information based on the information output from the detection device DD and the vehicle sensor 60. The detection information starts detection by the combination of the sensors (sensor configuration) in which the target is detected among the plurality of sensors included in the detection device DD and the vehicle sensor 60 and the combination of the sensors, and continuously from that time. The detection time, which is the time during which the detection is continued, and the detection distance determined based on the output and the detection time of each sensor in which the target is detected are included. The detection distance is determined during travel based on the results of pre-performed measurements that detect the target by a combination of the sensors in a situation where the true value of the distance to the target is known.

Hereinafter, the operation of the vehicle control device 100 configured as described above will be described.
(Explanation of principle)
First, the problems of the prior art will be described.
5 to 8 are diagrams for explaining the problems of the prior art.
<Issue 1>
When determining an action plan using a sensor, there is a problem (problem 1) that the recognition distance is insufficient and the recognition distance cannot be extended while maintaining high reliability of the sensor.

FIG. 5 is a diagram showing a target speed value curve when decelerating to a target stop position. FIG. 5A is a diagram showing a target speed value curve when decelerating at 0.6 G, in which the horizontal axis shows the vehicle position and the vertical axis shows the speed. The thick solid line in FIG. 5 (a) is the target speed value curve calculated when decelerating at 0.6 G, and the thick dashed line in FIG. 5 (a) starts shifting from a speed of 130 kph (error detection starts) and shifts at a speed of 50 kph. It is a target velocity value curve when (tolerance end). The vehicle position at speed 0 is the target stop position.

FIG. 5B shows a detection rate P which is a detection probability of a sensor (for example, a vehicle sensor such as a vehicle speed sensor, a detection device DD such as a radar or a camera) when determining the target speed value of FIG. 5A. (D | E) is shown, and FIG. 5 (c) shows the false detection rate P (D | E￣) of the sensor when determining the target speed value of FIG. 5 (a). Here, E represents a state in which an obstacle exists, and E￣
Represents a state in which no obstacle exists. Further, D represents an event of detecting an obstacle.
As shown in FIGS. 5 (b) and 5 (c), the false detection rate P (D | E￣) tends to decrease as the detection rate P (D | E) increases.
FIG. 5D is a diagram showing an example of a deceleration profile of the action plan in the case of automatic driving, and is determined by a predetermined algorithm in consideration of the detection rate and the false detection rate.

6 and 7 are diagrams for explaining an action plan based on the reliability of the sensor in determining the target velocity value curve of FIG. FIG. 6 shows an example in which the detection value of the sensor is set on the over-detection side in which the detection value of the sensor is used up to the region of low reliability in the determination of the action plan of FIG. FIG. 7 shows an example in which the detection value of the sensor is set to be used in a highly reliable region in the determination of the action plan of FIG.
The horizontal axis of FIGS. 6 and 7 indicates the vehicle position, where D2 is about 120 to 160 m, D1.5 is about 60 to 100 m, and D1 is about 30 to 50 m. The vertical axis of FIGS. 6 and 7 indicates the target speed of the vehicle, V1.5 is about 50 to 70 kph, and V2 is about 120 to 140 kph.

As shown in FIG. 6, a case where the driving environment condition of the automatic driving system is in the ODD (Operating Design Domain) and the driving can be performed at the speed V2 will be described. In the determination of the action plan in FIG. 5, if the detection value of the sensor is set on the over-detection side that uses the area with low reliability, the detection distance (distance of Gen2 ← Gen1) is long, but it is surrounded by the broken line frame a in FIG. As shown, the false detection rate of the sensor is large, so that false braking increases. In this way, if the region with low reliability is used as it is, the commercial value is lowered due to the increase in erroneous braking.

On the other hand, if the specification is set to apply the brake in the region where the reliability of the sensor detection value is high, the detection distance becomes short as shown by the x mark b in FIG. 7, and V2 that is allowed by ODD is not reached. In this way, if a highly reliable region is determined and used, it is necessary to reduce the cruising speed, and V2 ODD cannot be realized.
As described above, there is a problem that the recognition distance is insufficient and the recognition distance cannot be extended while maintaining high reliability of the sensor.

FIG. 8 is a diagram for explaining a direction for solving the problems in FIGS. 6 and 7.
By starting deceleration from a distant low-reliability position within a range that does not affect the ride comfort, it is possible to reliably stop in the high-reliability section. As shown by the arrow c in FIG. 8, the brake is gradually strengthened according to the reliability.
Actively use continuous "reliability" in action plans. That is, by extending the stopping distance, braking may be started even in a place with low reliability, and V2 may be achieved.

The present invention provides a method for calculating reliability and a method for deriving a specific method for linking action plans in order to solve <Problem 1>.

<Problem 2>
There is a problem (problem 2) that the level of safety is clarified, the reliability (accuracy, detection rate) of the sensor, and the accuracy of the action plan cannot be quantified.
In order to solve <Problem 2>, the present invention provides a formula that can quantitatively evaluate the reliability (accuracy, detection rate) of the sensor and its action plan, and clarifies the area that can be said to be safe.

[basic way of thinking]
The present invention introduces a "collision rate" into the action plan. That is, it is considered that the action plan has a "collision rate" in the same way that the recognition has a "recognition rate", and the "collision rate" is introduced into the action plan.

FIG. 9 is an explanatory diagram for explaining a collision rate for decelerating to a front vehicle position (target stop position) to prevent a collision. FIG. 9A shows a collision rate map that decelerates to the position of the vehicle in front to prevent a collision. In FIG. 9A, the horizontal axis is the position and the vertical axis is the velocity. FIG. 9 (b), the collision rate P shown in FIG. 9 (a) | expressed in gradation represent (C _{S t),} FIG. 9 (c), the peak of the collision rate when a rate of 0 shown in FIG. 9 (a) Shows the image of.

Figure 9 (a) to the indicated collision rate P (C | S _t), the detection rate and accuracy of the sensor, the action plan algorithm, and based on the accuracy of the action plan is calculated. The broken line box d in FIG. 9A can be said to be “no collision is possible” if the vehicle travels at a position and speed within this range.
The broken line box e in FIG. 9B shows the probability of ^{10-7 that "a collision is unthinkable".}
As shown in FIG. 9C, the collision rate P (C | _St ) at a speed of 0 is determined by the error standard deviation σ (not shown) of the sensor. If they are separated by 6σ, the collision rate has a probability of ^10-7.

According to the present invention, the accuracy of the sensor, the detection rate, and the algorithm of the action plan can be estimated in an integrated manner, and the quality of the algorithm can be determined.

FIG. 10A is a diagram illustrating a basic concept of the present invention.
[Related technologies such as collision rate]
<Collision rate parameters>
FIG. 10A shows a target speed value curve (thick solid line) of the vehicle 1 when the vehicle 1 decelerates to the position of the vehicle in front (target stop position D) to prevent a collision. The horizontal axis of FIG. 10A shows the vehicle position, and the vertical axis shows the speed. The vehicle 1 cannot always stop at the target stop position D due to various factors, and actually takes the position and speed shown by the dashed line shaded area in FIG. 10A.

As shown on the right side of FIG. 10A, the collision rate f indicating the possibility that the vehicle 1 collides with the vehicle in front is determined by the following three parameters. That is, the command value achievement probability density distribution 1001, the command value offset amount 1002, and the probability distribution 1003 based on the fusion accuracy reliability. The command value achievement probability density distribution 1001 and the command value offset amount 1002 are parameters related to the action plan in which the collision rate is introduced, and the probability distribution 1003 based on the fusion accuracy reliability is a parameter representing the recognition accuracy of the sensor. .. The outline of each parameter will be described below.

The command value achievement probability density distribution 1001 is one of the three parameters that determine the collision rate. The command value achievement probability density distribution 1001 represents the magnitude and certainty of the difference from the position where the own vehicle 1 actually reaches with respect to the command value of the traveling control unit 160 (vehicle behavior control unit).

The offset amount 1002 of the command value is one of the three parameters that determine the collision rate.
The offset amount 1002 of the command value is a parameter that determines how much forward the target stop position should be offset with respect to the vehicle in front.

The probability distribution 1003 based on the fusion accuracy reliability depends on the combination of sensors such as a camera, radar, and lidar, and the detection time and detection distance. The probability distribution 1003 based on the fusion accuracy reliability includes sensor redundancy, AND detection (see FIGS. 18 (a) and 27 below), OR detection (see FIGS. 18 (b) and 28 below), and HALF AND detection described later. It also depends on the distance detection mode such as (see FIGS. 18 (c) and 29 below).

Here, with reference to FIGS. 10A, 10B, 10C, and 12, the command value achievement probability density distribution 1001, the fusion accuracy reliability, the probability distribution 1003 based on the fusion accuracy reliability, the collision rate f, and the offset of the command value. The relationship of the quantity 1002 will be described in detail.

Command value achievement probability density distribution

The command value achievement probability density distribution 1001 represents the distribution of the position where the vehicle 1 actually stopped when a certain deceleration is commanded to the vehicle 1 with respect to the target stop position. The command value achievement probability density distribution 1001 shown in FIG. 10A shows the distribution of the stop position of the vehicle 1 when the vehicle 1 tries to stop at the position D. The command value achievement probability density distribution 1001 is a characteristic that is measured in advance through a process of measuring the stop position when a certain deceleration is commanded to the vehicle 1.

It is assumed that the action plan generation unit 200 determines the maximum allowable deceleration in the action plan according to the mode of automatic operation and the like. Of the command value achievement probability density distributions measured in advance in the above process, those traveling at the same or substantially the same deceleration as the maximum deceleration are regarded as the command value achievement probability density distribution 1001 in the action plan to be executed from now on.

fusion accuracy reliability

Fusion accuracy The reliability is the time that detection is started by the combination of sensors (sensor configuration) in which a target is detected among multiple sensors and the combination of the sensors, and the detection is continuously continued from that time. It is calculated based on a certain detection time. In general, the greater the number of sensors that detect a target, or the longer the detection continues, the higher the fusion accuracy and reliability. Fusion accuracy The reliability varies depending on the combination of sensors that detect the target.

FIG. 10B shows an example of the combination of three types of sensors, Camera, Lidar, and Radar, and the fusion accuracy reliability with respect to the detection time. The vertical axis of FIG. 10B is arranged so that the combination of the sensors that have detected the target is higher in reliability, and the horizontal axis represents the time when the target is continuously detected by the sensor.

For example, when it is detected by the sensor configuration a (Camera) and the detection time is D, the fusion accuracy reliability is Low (low), and when it is detected by the sensor configuration f (Lida + Camera) and the detection time is B, it is low. , The fusion accuracy reliability is detected by Mid (medium) and the sensor configuration g (Radar + Lidar + Camera), and when the detection time is D, the fusion accuracy reliability is High (high). The fusion accuracy reliability may be obtained by dividing the positive detection rate for each detection time measured for each sensor configuration into three stages of Low, Mid, and High based on a preset threshold value, and positive detection. The rate may be treated as a numerical value as it is.

Probability distribution based on fusion accuracy reliability

The probability distribution based on the fusion accuracy reliability indicates how the detection distance measured based on a plurality of sensors is distributed with respect to the true value. FIG. 10C shows the relationship between the fusion accuracy reliability, the width of the probability distribution 1003 based on the fusion accuracy reliability, the command value achievement probability density distribution 1001, the collision rate f, and the offset amount 1002.

For example, when the fusion accuracy reliability is focused on any of Mid and the like, when the variation with respect to the true value of the detection distance is large, the width of the probability distribution 1003 based on the fusion accuracy reliability is widened, and the width of the probability distribution 1003 is widened with respect to the true value of the detection distance. When the variation is small, the width of the probability distribution 1003 based on the fusion accuracy reliability becomes narrow. The probability distribution 1003 based on the fusion accuracy reliability and its width can be obtained in advance through a process of testing what happens to the distribution of the error of the detection distance in a situation where the true value of the distance is known.

Focusing on the case where the width of the probability distribution 1003 based on the fusion accuracy reliability is "medium", when the fusion accuracy reliability is High, the height of the probability distribution 1003 based on the fusion accuracy reliability becomes high. When the fusion accuracy reliability is Low, the height of the probability distribution 1003 based on the fusion accuracy reliability becomes low. The height of the probability distribution 1003 based on the fusion accuracy reliability is determined so that the area of the distribution is substantially proportional to the detection rate of the sensor. The width and height of the probability distribution 1003 based on the fusion accuracy reliability are obtained in advance through the process of obtaining the detection rate characteristics, which tests whether or not the detection target is detected in the situation where the presence or absence of the detection target is known. It is a characteristic to keep.

Offset amount of command value

The collision rate f, which indicates the possibility that the vehicle 1 collides with the vehicle in front, is determined by the relative positional relationship between the command value achievement probability density distribution 1001 and the probability distribution 1003 based on the fusion accuracy reliability, and is an offset of the command value. The quantity 1002 defines this relative positional relationship.
That is, the offset amount 1002 of the command value indicates the difference between the positions of the vertices of the two distributions, the command value achievement probability density distribution 1001 and the probability distribution 1003 based on the fusion accuracy reliability, and the collision rate f is increased by this difference. Determined

In other words, the offset amount 1002 of the command value can be set so as to reach the target collision rate (target collision rate). The collision rate f is obtained by convolution when the command value achievement probability density distribution 1001 and the probability distribution 1003 based on the fusion accuracy reliability are shifted by the offset amount 1002 of the command value. The value offset amount 1002 can be determined. The nine graphs shown in FIG. 10C show the offset amounts having the same collision rate when the probability distribution 1003 based on the fusion accuracy reliability and the fusion accuracy reliability is given to a certain command value achievement probability density distribution 1001. It expresses how 1002 changes. The target collision rate is a fixed value that is determined in advance as a design, and is a value that is determined by the detection rate of the detected object, the severity of collision with the detected object, and the like.

Focusing on the case where the fusion accuracy reliability is Mid or the like in FIG. 10C, when the width of the probability distribution 1003 based on the Fusion accuracy reliability is narrow, the offset amount 1002 of the command value becomes small, and the Fusion accuracy reliability. When the width of the probability distribution 1003 based on is wide, the offset amount 1002 of the command value becomes large. Focusing on the case where the width of the probability distribution 1003 based on the fusion accuracy reliability is medium, when the fusion accuracy reliability is High, the offset amount 1002 of the command value becomes large, and when the fusion accuracy reliability is Low, the command is issued. The value offset amount 1002 becomes smaller.

Width of probability distribution based on fusion accuracy reliability

The width of the probability distribution 1003 based on the fusion accuracy reliability varies depending on the sensor configuration (combination with the type of sensor), the detection time, the distance to the target, and the like. Generally, when the distance to the target is long, the width of the probability distribution 1003 based on the fusion accuracy reliability is "wide", and when it is close, it becomes "narrow". Further, when comparing the distances to the same target, for example, the sensor configuration a (Camera) is "wide", the sensor configuration c (Radar) is "medium", and the sensor configuration d (Radar + Radar) is "narrow". .. Also, in general, the longer the detection time, the narrower the detection time.

Estimating the width of the probability distribution based on fusion accuracy reliability during driving

As described above, the width of the probability distribution 1003 based on the fusion accuracy reliability can be obtained in advance through the process of testing what the distribution of the error of the detection distance will be in the situation where the true value of the distance is known. can. That is, the width of the probability distribution 1003 based on the fusion accuracy reliability can be obtained for each combination of the sensor configuration (combination with the type of sensor), the detection time, and the detection distance. Then, a table (detection probability distribution width table) (not shown) is configured in which the sensor configuration (combined with the sensor type), the detection time, and the detection distance are input and the width of the probability distribution 1003 based on the fusion accuracy reliability is output. It can be stored in a storage device (Read Only Memory (ROM) or the like) of the vehicle control device. Thereby, the width of the probability distribution 1003 based on the current fusion accuracy reliability can be estimated by referring to the detection probability distribution width table from the combination of the current sensor configuration, the detection time, and the detection distance during traveling.

The above-mentioned detection probability distribution width table has been described as acquiring the width of the probability distribution 1003 based on the fusion accuracy reliability from the three parameters of the sensor configuration, the detection time, and the detection distance. The input information to the table may be reduced by using two parameters of degree (High / Mid / Low) and detection distance as input parameters, or using two parameters of sensor configuration and detection time as input parameters. Further, in order to reduce the input information, the accuracy (quantization width) for each parameter can be adjusted in addition to the number of parameters. These choices are a trade-off between accuracy and cost.

Offset amount calculation procedure while driving

FIG. 12 shows how to obtain the offset amount 1002 by the sensor configuration and the detection time when the command value achievement probability density distribution 1001 and the target collision rate are given to the target at a certain distance while traveling. It is a figure explaining.

In FIG. 12, the horizontal axis represents the width of the probability distribution based on the fusion accuracy reliability, and the vertical axis represents the offset amount. In FIG. 12, the given command value achievement probability density distribution 1001 and the target collision rate are based on the probability distribution 1003 based on the fusion accuracy reliability actually measured for the combination of the sensor configuration and the detection time shown in FIG. 10B. The offset amount 1002 obtained is plotted with the width of the probability distribution 1003 based on the fusion accuracy reliability on the horizontal axis.

In FIG. 12, the plots when the fusion accuracy reliability is High, Mid, and Low are surrounded by broken line regions High, Mid, and Low, respectively. In FIG. 12, for example, a combination in which a target is detected by the sensor configuration e (Radar + Camera) and the detection time is B is indicated by reference numerals eB in FIG. Plots for other combinations are also indicated by similar signs.

As shown in FIG. 12, the offset amount has a linear characteristic having a different slope depending on the fusion accuracy reliability with respect to the width of the probability distribution 1003 based on the fusion accuracy reliability. fusion accuracy The higher the reliability, the greater the slope of the characteristics.

Since the command value achievement probability density distribution 1001 also differs depending on the commanded deceleration, the characteristics shown in FIG. 12 also change depending on the deceleration, but show substantially the same tendency. Therefore, by storing the above linear characteristic parameters (slope, etc.) for each deceleration and each fusion accuracy reliability in a table (offset amount characteristic parameter table) (not shown), a simple calculation can be performed while driving. The offset amount can be calculated.

Specifically, the offset amount characteristic parameter table is referred to according to the maximum deceleration in the current action plan and the fusion accuracy reliability (High / Mid / Low), and one of the linear characteristics as shown in FIG. 12 is selected. Then, the offset amount 1002 can be calculated by applying the width of the probability distribution 1003 based on the fusion accuracy reliability obtained by referring to the detection probability distribution width table to the selected linear characteristics.

In the above calculation method, the offset amount 1002 was calculated by applying the width of the probability distribution 1003 based on the separately estimated fusion accuracy reliability to one of the linear characteristics shown in FIG. On the other hand, as a general form, a table may be provided in which the deceleration, the sensor configuration, the detection time, and the detection distance are input and the output is the offset amount, and the offset amount 1002 may be obtained only by referring to the table once.

As understood in the above description, the command value achievement probability density distribution 1001 can be estimated based on the deceleration. Further, the probability distribution 1003 based on the fusion accuracy reliability can be estimated based on the sensor configuration, the detection time, and the detection distance. Then, the offset amount 1002 that gives the target collision rate can be calculated from the width of the estimated command value achievement probability density distribution 1001 and the probability distribution 1003 based on the fusion accuracy reliability. Then, an offset amount table (not shown) in which the deceleration, the sensor configuration, the detection time, and the detection distance are input and the output is the offset amount 1002 can be configured by using Read Only Memory (ROM) or the like.

However, in the case of this configuration, the capacity of the offset amount table increases or decreases depending on the deceleration given as input to the offset amount table, the number of sensor combinations, the quantization width of the detection time, the quantization width of the detection distance, etc., and the accuracy is improved. The capacity increases as it increases. That is, as a method of acquiring the offset amount 1002, whether to calculate based on the linear characteristics as shown in FIG. 12 or to use a table reference is in a trade-off relationship between accuracy, calculation time, and cost.

Returning to FIG. 10A, an example of the flow of calculating the offset amount 1002 in the vehicle control device 100 will be described below.

As shown in FIG. 10A, the probability distribution 1003 based on the fusion accuracy reliability is distributed centering on the vehicle in front. At this time, it is assumed that the own vehicle 1 tries to stop at the position of reference numeral D in FIG. 10A to prevent a collision. The command value achievement probability density distribution 1001 at this time is a distribution centered on the position D as shown in FIG. 10A. In the example shown in FIG. 10A, the command value achievement probability density distribution 1001 and the probability distribution 1003 based on the fusion accuracy reliability partially overlap. As described above, the area of this overlap represents the collision rate in this stop operation. Further, this overlap is changed by changing the offset amount 1002, and therefore the collision rate also changes.

Since the own vehicle 1 in motion cannot know the actual position of the vehicle in front, it cannot know the position of the central axis of the probability distribution 1003 based on the fusion accuracy reliability. However, as described above, the time during which detection is started by the combination of the sensors in which the target is detected among the plurality of sensors (sensor configuration) and the combination of the sensors and the detection is continuously continued from that time. The width of the probability distribution 1003 based on the fusion accuracy reliability can be estimated by obtaining the fusion accuracy reliability from the combination with the detection time and referring to the pre-measured data. Further, as described above, the command value achievement probability density distribution 1001 in the current stop operation can be estimated by referring to the pre-measured data based on the currently set maximum deceleration.

Then, on the premise of the command value achievement probability density distribution 1001 estimated in this stop operation, as explained with reference to FIG. 12, the width of the probability distribution 1003 based on the fusion accuracy reliability and the fusion accuracy reliability is determined in advance. By applying to the measured data, it is possible to determine the offset amount 1002 that gives a certain target collision rate.

It should be noted that the command value is such that the collision rate obtained by convolution when the command value achievement probability density distribution 1001 and the probability distribution 1003 based on the fusion accuracy reliability are shifted by the offset amount 1002 of the command value matches the target collision rate. It was explained that the offset amount 1002 of the above is determined. However, as described above, an offset amount table in which the deceleration, the sensor configuration, the detection time, and the detection distance are input and the output is the offset amount is stored in the Read Only Memory (ROM), and this table is referred to when traveling. Then, since the information of the command value achievement probability density distribution 1001 and the probability distribution 1003 based on the fusion accuracy reliability is already expressed in this table, it is not necessary to perform the calculation for convolution during traveling. Therefore, it is not necessary to estimate or use the command value achievement probability density distribution 1001 and the probability distribution 1003 itself based on the fusion accuracy reliability in the calculation.

Here, as described above, the reason for obtaining the command value achievement probability density distribution 1001 based on the currently set maximum deceleration will be described. As the deceleration increases, the command value achievement probability density distribution 1001 expands, and the set offset amount increases. By assuming the maximum deceleration in the action plan, a safe offset amount suitable for the action plan can be obtained.
The above operation will be described below with reference to FIG.

<Flow of determining the command value of the action plan>
FIG. 11 is a flow for determining the offset amount of the command value of the action plan based on the characteristics shown in FIG. An example of an embodiment in which the offset amount 1002 of the command value is determined based on the characteristics shown in FIG. 12 will be described below. The offset amount 1002 of the command value defines the target stop position for stopping the vehicle 1 as an offset from the position indicated by the detection distance of the distance detection unit in the action plan generated by the action plan generation unit 200. ..

STEP1;
The target collision rate setting unit 210 sets the target collision rate. The target collision rate is a predetermined fixed value that is determined in advance as a design, and is a value that is determined in consideration of the detection rate of the detected object, the severity of collision with the detected object, and the like.
STEP2;
The fusion accuracy reliability estimation unit 240 determines the fusion accuracy reliability based on the current sensor configuration and the detection time. The Fusion accuracy reliability is determined based on the positive detection rate measured in advance for the combination of sensors, as already described with reference to FIG. 10B. The fusion accuracy and reliability is the time that detection is started by the combination of sensors (sensor configuration) in which a target is detected among multiple sensors and the combination of the sensors, and the detection is continuously continued from that time. It is calculated based on a certain detection time. In general, the larger the number of sensors that detect a target, or the longer the target is continuously detected, the higher the fusion accuracy and reliability. fusion accuracy The reliability varies depending on the combination of sensors that detect the target.

STEP3;
The command value achievement probability density distribution estimation unit 220 and the command value offset amount calculation unit 230 determine command value table (probability density, offset amount) candidates. The command value achievement probability density distribution estimation unit 220 determines a candidate for the command value achievement probability density distribution 1001 from the currently set maximum deceleration. As described above, the candidates for the command value achievement probability density distribution 1001 are determined based on the characteristics obtained by the measurement in advance through the prior process of measuring the stop position when a certain deceleration is commanded.

The command value offset amount calculation unit 230 determines a candidate for the command value offset amount based on the fusion accuracy reliability determined by the fusion accuracy reliability estimation unit 240 and the candidates for the command value achievement probability density distribution 1001.

This step corresponds to selecting one of the dashed area High, Mid, and Low in FIG. For example, when the fusion accuracy reliability is Mid, the offset amount corresponding to the broken line region Mid in FIG. 12 is a candidate.

STEP4;
The fusion accuracy reliability estimation unit 240 estimates the width of the probability distribution based on the fusion accuracy reliability. The command value offset amount calculation unit 230 selects one from the offset amount candidates selected in STEP 3 based on the width of the probability distribution 1003 based on the estimated fusion accuracy reliability. This is detected by, for example, the sensor configuration e (Radar + Camera), and when the detection time is B, the fusion accuracy reliability is determined to be Mid, and the characteristics when the fusion accuracy reliability is Mid, as shown in FIG. As a result of fitting the width of the probability distribution 1003 based on the estimated Fusion accuracy reliability, it corresponds to selecting the offset amount 1002 corresponding to the reference numerals EB in FIG.

As described above, the width of the probability distribution 1003 based on the fusion accuracy reliability is, for example, a detection probability distribution width table that stores data of the width of the probability distribution obtained by measuring the current sensor configuration and the detection time in advance. Obtained by inputting to (not shown) and obtaining table output.
STEP5;
The action plan generation unit 200 determines the command value table using the offset amount 1002 obtained in STEP4. As a result, the target position where the action plan generation unit 200 stops the vehicle 1 is determined.

The flow of STEP1 to STEP4 described above is an example and is not limited thereto. For example, in the determination of the offset amount by STEP3 and STEP4, as described above, the offset amount table (not shown) in which the deceleration, the sensor configuration, the detection time, and the detection distance are input and the output is the offset amount is used only memory. It can be executed by storing it in (ROM) and referencing it by the action plan generation unit 200. In this embodiment, the width of the command value achievement probability density distribution 1001 and the probability distribution 1003 based on the fusion accuracy reliability is reflected in the data stored in the offset amount table, and the command value achievement probability density distribution 1001 and It is not necessary to individually obtain the width of the probability distribution 1003 based on the fusion accuracy reliability during traveling. The embodiments described below will be described on the premise that the offset amount that gives the target collision rate from the deceleration, the sensor configuration, the detection time, and the detection distance can be obtained by a method similar to the method using the offset amount table described above. do.

<Collision rate>
The collision rate p (C | _St ) represented by the collision rate map is calculated according to the calculation model shown by the equation (1). The details of the calculation process related to this calculation will be described later.

Ｓ_ｔ： 時刻ｔでの状態（位置、速度）
Ｓ_ｔ＋１： 時刻ｔ＋１での状態（位置、速度）＝次の状態
ｐ（Ｃ｜Ｓ_ｔ＋１）：次の状態（位置、速度）におけるコリジョン率
ｐ（Ｓ_ｔ＋１｜α_Ｄ（Ｓ_ｔ，ｄ＾））：状態Ｓ_ｔに存在するときにｄ＾を観測したときに発生するアクションα_Ｄによって次の状態Ｓ_ｔ＋１に遷移する確率
ｐ（ｄ＾｜ｄ）：物標状態がｄであるときにｄ＾を観測する確率

_St : State at time t (position, speed)
_{St + 1} : State at time t + 1 (position, velocity) = next state p (C | _{St + 1} ): collision rate in the next state (position, velocity) p ( _{St + 1} | α _D ( _St , d ^)) ): state S probability transitions to the next state S _{t + 1} by the action alpha _D generated when observing d ^ when present in _t p (d ^ | d): d when a target state is d Probability of observing ^

<Flowchart>
FIG. 13 is a flowchart of vehicle control of the vehicle control device 100 of the embodiment.

This flow is repeatedly executed by the automatic operation control unit 120 (see FIG. 2) at a predetermined cycle.

Step S11
The automatic driving control unit 120 (distance detection unit) detects the distance to an obstacle to be avoided based on the output of the detection device DD (distance detection unit).

Step S12
The collision rate map setting unit 250 detects the current target among a plurality of sensors of the predetermined target collision rate, the maximum deceleration currently set by the action plan generation unit 200, and the detection device DD of the distance detection unit. The collision rate map is determined based on the combination of the existing sensors (sensor configuration), the detection distance detected by the sensor, and the detection time continuously detected by the sensor. The target collision rate is a predetermined fixed value that is determined in advance as a design, and is a value that is determined in consideration of the detection rate of the detected object, the severity of collision with the detected object, and the like.

This collision rate map is a probability based on the fusion accuracy reliability estimated based on the command value achievement probability density distribution 1001 estimated from the maximum deceleration, the sensor configuration, the detection distance, and the detection time. It is created on the premise of an offset amount determined so that the collision rate calculated based on the overlap of the distribution 1003 is equal to the target collision rate. In other words, the collision rate map sets the collision rate at that position to be the target collision rate based on the target collision rate, maximum deceleration, sensor configuration, detection distance, and detection time. It is created on the premise of position.

The collision rate map obtained here is an approximation of the collision rate with respect to the positions and velocities corresponding to the discrete grid points. Each of the discrete grid points means a discrete "state".

Step S13
The action plan generation unit 200 formulates an action plan based on the collision rate in the current state (position, speed) of the own vehicle 1 on the collision rate map and the target collision rate.

Step S14
The travel control unit 160 controls at least the speed of the own vehicle 1 based on the action plan, and ends one process of this flow.

The automatic operation control unit 120 executes steps S11 to S14 every time a predetermined time elapses.

The features of the collision rate map used for vehicle control will be described below.

<Collision rate map>
FIG. 14 is a diagram showing a collision rate map 1000. FIG. 14A shows a collision rate map 1000 when decelerating to the position of the vehicle in front to prevent a collision. In FIG. 14A, the horizontal axis is the position and the vertical axis is the velocity. 14 (b) shows the collision rate represented in FIG. 14 (a) in gradation as a reliability rate in observation. FIG. 14 (c) shows an image of the peak of the collision rate at a speed of 0 in the collision rate map 1000 of FIG. 14 (a).

The collision rate map 1000 shown in FIG. 14 is for explaining how an action plan can be created with reference to the collision rate map, and it is also assumed that the actual position of the vehicle in front is known. It is a theoretical one created in.

Once the detection rate and accuracy of the sensor, the accuracy of the action plan, and the collision rate of the final state are determined, the collision rate for each state consisting of the speed of the vehicle and the position of the vehicle with respect to the target can be calculated. Based on the calculated collision rate, the collision rate map 1000 shown in FIG. 14A is created.

In the collision rate (0 to 1) shown in FIG. 14 (b), for example, a human collision rate or a value smaller than that is used as the target collision rate (see reference numeral i in FIG. 14 (b)). Here, the "human collision rate" means the "collision rate that occurs when a person (driver) drives the vehicle 1 in the same road environment". Setting the collision rate as the target collision rate compares the case where the autonomous driving is driven and the case where a person is driving, and in order for the automatic driving to be safer, the collision rate is It is based on the idea that it should be lower than the collision rate that occurs when a person drives.

By connecting the states (positions, speeds) at which the collision rate is the target collision rate on the collision rate map 1000 of FIG. 14A, it is possible to obtain a target speed value curve for traveling at the target collision rate. can. The arrow (→) in the collision rate map shown in FIG. 14A is the target collision rate.

The collision rate at a velocity of 0 corresponds to the probability distribution 1003 based on the fusion accuracy reliability, and has a normal distribution as shown in FIG. 14 (c). Assuming that the standard deviation of this normal distribution is σ, the collision rate has a probability of ^{10-7 at a position 6σ away, and no collision is considered.}

An object of the present invention is to achieve both a safety factor and comfort. For this purpose, an appropriate action plan is set within the area of the collision rate map 1000 shown in FIG. 14 (a).

When the collision rate is lower than the first threshold value lower than the target collision rate (see FIG. 14 (a) [collision rate low region]), braking is tolerated, that is, the set speed is maintained as much as possible. Comfort is important, and although it is not jerky, it will allow sudden braking.

When the collision rate is lower than the target collision rate and equal to or higher than the second threshold value which is equal to or higher than the first threshold value (see FIG. 14 (a) [collision rate high region]), braking is performed diligently. Safety is important, and although it is jerky, it does not cause sudden braking. FIG. 14A shows an example in which the second threshold value is equal to the first threshold value.

The representation of the shade of the collision rate map 1000 shown in FIG. 14A represents an index for controlling so as not to exceed the collision rate designed in advance. Further, on the control side, when performing automatic operation control, it is possible to control so as not to fall into a region where the shading is dark.

<Action plan algorithm and accuracy>
The algorithm and accuracy of the action plan will be explained.
Finding the certainty that the purpose of the action plan algorithm is achieved corresponds to finding p ( _{St + 1} | α _D ( _St , d ^)) in Eq. (1).

FIG. 15 is a diagram illustrating the relationship between the action plan and the risk of collision. The same parts as those in FIG. 10A are designated by the same reference numerals.
The vehicle control device 100 (see FIG. 2) controls the vehicle 1 so as to stop at the position D in FIG. However, it is difficult to achieve a stop at position D due to several factors such as road surface conditions and vehicle body failure. Whether or not this can be achieved is indicated by the probability density distribution. In the case of this embodiment, the command value achievement probability density distribution estimation unit 220 (see FIG. 4) estimates the command value achievement probability density distribution 1001.

As shown in FIG. 15, the collision risk is indicated by the portion where the command value achievement probability density distribution 1001 overlaps the collision region 1200 (see reference numeral j in FIG. 15). If it is desired to be safer, this portion (see reference numeral j in FIG. 15) may be made smaller. As a method of making it smaller, it is possible to make the probability density distribution closer to the median value (to make the factors such as the road surface condition and the vehicle body failure smaller), but since this is difficult, the position D is set from the collision area 1200. It will be set to the front.

<Sensor reliability (detection rate, accuracy)>
The reliability (detection rate, accuracy) of the sensor will be described.
Obtaining the reliability of the sensor corresponds to obtaining p (d ^ | d) in the equation (1).

FIG. 16 is a diagram illustrating an error distribution of the sensor. The same parts as those in FIG. 15 are designated by the same reference numerals. The “sensor error distribution” shown at the bottom of FIG. 16 shows the distribution of errors that occur as a result of the sensor detecting a target. This distribution is centered on the position of the target. Further, this distribution is the probability distribution 1003 based on the above-mentioned Fusion accuracy reliability.
As shown in FIG. 16, by providing a margin M at the braking position, the probability of collision can be reduced.

<Effects of action plans and sensors on collision risk>
Explain the impact of action plans and sensors on collision risk.
FIG. 17 is a diagram illustrating the influence of the action plan and the sensor on the collision risk. FIG. 17A is a collision rate map when the maximum braking is −0.6 G and the error standard deviation of the sensor is σ1. The white solid line in FIG. 17A shows an example of the target collision rate.

FIG. 17B shows the target collision rate when the action plan setting of the collision rate map of FIG. 17A is changed. Increasing the maximum braking allowed in the action plan increases the vehicle 1's ability to deal with obstacles and reduces the collision rate. In this case, the combination of the position having the target collision rate and the speed on the collision rate map moves to the upper right (target side, high speed side). That is, the region where the collision rate is low spreads. As an example, as shown by the arrow k in FIG. 17 (b), if the action plan changed to the maximum braking -1.0 G is set, the deceleration G that can be used for braking can be increased to expand the state where the collision risk is low. Can be done. Therefore, the timing of braking can be delayed to the very limit.

FIG. 17 (c) shows how the collision rate map of FIG. 17 (a) changes when the sensor performance deteriorates. When the error standard deviation of the sensor becomes σ2 (σ1 <σ2) and the sensor performance deteriorates, the collision rate is estimated to be high in this embodiment, and the collision rate low region in the collision rate map becomes narrow. This is represented by the arrow l in FIG. 17 (c). In this case, the risk of collision can be reduced by generating an action plan so as to reduce the traveling speed.

<How to use the reliability (detection rate) of multiple sensors>
How to use the reliability (detection rate) of multiple sensors will be described.
FIG. 18 is a diagram illustrating how to use the reliability (detection rate) of the plurality of sensors. FIG. 18 (a) shows an example of detecting a plurality of sensors by AND, FIG. 18 (b) shows an example of detecting a plurality of sensors by OR, and FIG. 18 (c) shows an example of detecting a plurality of sensors by OR. An example of braking according to the detection state is shown.

Let the plurality of sensors be sensor 1 and sensor 2. Here, D ₁ represents an event in which the target is detected by the sensor 1, D ₂ represents an event in which the target is detected by the sensor 2, and E represents a state in which the target exists. The probability that the sensor 1 detects the target in the presence of the target _{is expressed as p (D 1} | E). The probability that the sensor 2 detects the target in the presence of the target _{is expressed as p (D 2} | E). The probability that both the sensor 1 and the sensor 2 detect the target in the presence of the target _{is expressed as p (D 1} ∩ _{D 2} | E). The probability of an event in which the sensor 1 detects the target and the sensor 2 does not detect the target in the presence of the target _{is expressed as p (D 1} ∩ _{D 2} ￣ | E). The probability of an event in which the sensor 1 does not detect the target and the sensor 2 detects the target in the presence of the target _{is expressed as p (D 1} ∩ _{D 2} | E). The probability of an event in which both sensor 1 and sensor 2 do not detect a target in the presence of a target _{is expressed as p (D 1} ∩ _{D 2} ￣ | E). _{Regarding the probability p (D 1} ∩ _{D 2} | E) detected by both sensors at the same time, when the equation (2) holds, it is said that the event D ₁ | E and the event D ₂ | E are not independent.

p (D ₁ ∩ _{D 2} | E) ≠ p (D ₁ | E) p (D ₂ | E)… (2)

In the example detected by AND in FIG. 18A, braking with a deceleration of 0.6 G is performed at a ratio of 0.4 to 1.0 as a whole, and cruising is performed at a ratio of 0.6. When detecting with AND, braking is performed only when both sensors detect a target at the same time. The probability of this event in the presence of a target is p (D ₁ ∩ _{D 2} | E). If either or both sensors do not detect the target, cruising is performed. The probability of an event in which both or either sensor does not detect a target in the presence of a target is p (D ₁ ∩ _{D 2} ￣ | E) + p (D _{1 ￣ ∩ D 2} | E) + p (D ₁ ￣ ∩ D ₂ ￣ | E).
An example of an action plan using AND detection will be described later with reference to FIG. 27.

In the example detected by OR in FIG. 18B, braking with a deceleration of 0.6 G is performed at a ratio of 0.9 to 1.0 as a whole, and cruising is 0 to 1.0 as a whole. It is carried out at a rate of 1. When detecting by OR, braking is performed when the target is detected by either or both sensors. The probabilities of events where a target is detected by either or both sensors in the presence of the target are p (D ₁ ∩ _{D 2} | E), p (D ₁ ∩ _{D 2} ￣ | E), p (D). ₁ ￣ ∩ D ₂ | E). If both sensors do not detect the target, cruising is carried out. The probability of this event in the presence of a target is p (D ₁ ∩ _{D 2} ￣ | E).
An example of an action plan using OR detection will be described later with reference to FIG. The superiority or inferiority or application conditions of the action plan using AND detection and the action plan using OR detection cannot be unequivocally determined.

When braking according to the detection state of the sensor in FIG. 18 (c), the ratio of brake braking of 0.6 G is 0.4 and its probability p (D ₁ ∩ _{D 2} | E) and 0.1 G. Brake braking rate 0.3 and its probability p (D ₁ ∩ _{D 2} ￣ | E) and 0.05 G brake braking rate 0.2 and its probability p (D ₁ ∩ _{D 2} | E) , A combination with a cruising ratio of 0.1 and its probability p (D ₁ ∩ _{D 2} ￣ | E) can be exemplified.
An example of an action plan for braking according to the detection state will be described later as HALF AND detection in FIG. The method of using the reliability (detection rate) of the plurality of sensors shown in FIG. 18 is an example.

As described above, in the present embodiment, the detection state of the two sensors is biased, and the braking strength is changed according to the detection rate of each sensor. Aim for the logic of the action plan that balances the above.

<Collision rate by algorithm α>
The changes in collision rate and discomfort rate due to algorithm α will be explained.
FIG. 19 is a state transition diagram illustrating changes in the collision rate and discomfort rate due to algorithm α.
The algorithm α surrounded by the broken line in FIG. 19 finds the action A _i when the total collision rate becomes a predetermined collision rate P _{c with α = {D → A i} , D￣ → A ₁ }. Here, D → A _i indicates that the action A _i is adopted when the target is detected, and D￣ → A _{1 indicates that the action A 1} is adopted when the target is not detected.
In FIG. 19 and each of the following equations, behavior: A _i , state: S _i , collision: C, discomfort: U, detection: D, non-detection D ￣, obstacle presence: E, no obstacle: E ￣ are shown.
The positive detection rate of the sensor: p (D | E) and the false detection rate of the sensor: p (D | E￣) are input to the algorithm α.

_{The collision rate p α} (C | E) due to the algorithm α is expressed by the equation (3), and the collision rate p (C | A _{i ) when the action A i} is selected is expressed by the equation (4).
The collision rate at the time of detection taking value closest to P _c action A _i is, collision rate by the Arigorizumu alpha p | collision rate in the case of selecting the (C E) an action A _i p (C | A _i ) And, it is expressed by the equation (5).
Equation (5) shows an algorithm α for finding _{an action A i in} which the total collision rate is the closest value to a _{predetermined collision rate P c.} Once the target collision rate P _c is determined, the policy of the algorithm α is determined.

Figure 19 is a state _{S init} is whether a transition to a state of the state _S 1 to S ₃ throat by action _A 1 to A ₃ probability _{p (S 1 | A 1)} , p (S 2 | A 1), p ( S ₃ | A ₁ ), p (S ₁ | A ₂ ), p (S ₂ | A ₂ ), p (S ₃ | A ₂ ), and p (S ₁ | A ₃ ), p (S ₂ | A ₃ ) and p (S ₃ | A ₃ ) are shown. In other words, the state S ₁ to S ₃ can be regarded as a state to which the transition from the state S _init by an event that occurs in the above probability.

By determining the above states S _{1 to} S _3, the collision probabilities p (C | S ₁ ), p (C | S ₂ ), p (C | S ₃ ) _{generated by the states S 1 to} S _{3 and} The discomfort rates p (U | S ₁ ), p (U | S ₂ ), and p (U | S ₃ ) generated by the states S _{1 to} S ₃ are determined.

FIG. 20 shows the probabilities p (S ₁ | A ₁ ), p (S ₂ | A ₁ ), and p (S ₃ ) of transitioning to the states S ₁ , S ₂ , and S _{3 according to the actions A 1 to} A _{3 shown in FIG.} | A ₁ ), p (S ₁ | A ₂ ), p (S ₂ | A ₂ ), p (S ₃ | A ₂ ), and p (S ₁ | A ₃ ), p (S ₂ | A ₃₎ ), P (S ₃ | A ₃ ) shows specific examples.

_{The collision rate p α} (C | E) due to the algorithm α is expressed by the equation (3), and the collision rate p (C | A _{i ) when the action A i} is selected is expressed by the equation (4).
_{Then, the action A i} at the time of detection in which the collision rate _{takes the value closest to P c is the collision rate p α} (C | E) due to the algorithm α and the collision rate p (C | A) when the action A _i is selected. _{Based on i} ), it is expressed by the equation (5).

The discomfort rate p (U | A _i ) when the action A _i is selected is expressed by the equation (6), and the discomfort rate p _α (C | E￣) of the algorithm α is expressed by the equation (7).

<Formation of state transition tree for behavior>
The evaluation of the collision rate by the algorithm in the network by forming the state transition tree for the behavior will be described.
First, the formation of a state transition tree for actions will be described.
Figure 20 shows an example of state transition to the state S ₁ to S ₃ by acting in a certain state. A state transition tree is formed by performing such a state transition in each state.

FIG. 21 is a diagram illustrating a state transition tree for actions. The horizontal axis X in FIG. 21 indicates the position, and the vertical axis V indicates the speed. Each state on the state transition tree has a collision rate, and transitions to another state until it reaches the target stop position (collision position) while decelerating or cruising, or until it avoids a collision. Therefore, in the state transition tree, the states are described so as to transition to the right and downward.
The ● mark in FIG. 21 indicates the state, and the broken line arrow in FIG. 21 indicates that the state branches and transitions as the position moves. Since the distance to the target stop position (collision position) decreases as the vehicle moves in each stage (row direction) of the state, the detection rate of the sensor (including the grasp rate of the road surface condition) increases.
Further, in FIGS. 22 and 23, the thick arrow indicates the transition of interest, and the thin arrow indicates the transition after the next stage of the transition of the thick arrow.

For example, in the case of FIG. 21, in the state shown by the broken line box m in FIG. 21, the collision rate is 1 because the position of the target is reached and the speed is high. In the state shown by the broken line box n in FIG. 21, the speed becomes 0 (stopping at speed 0) before reaching the collision position, so that the collision rate = 0 in collision avoidance. In the state shown by the broken line box o in FIG. 21, the collision rate is 0.5 because the speed is 0 although it is the collision position.

Next, the change in the detection rate on the state transition tree with respect to the behavior of FIG. 21 will be described.
FIG. 22 is a diagram illustrating a change in the detection rate on the state transition tree with respect to the behavior of FIG. 21. The detection rate of the sensor improves as it approaches the target stop position (collision position) for each stage (row direction) of the state, and 0.2, 0.4, 0.6, 0.8, 1.0, It is assumed to be 1.0.
FIG. 22A shows an example of a transition from the initial state in the state transition tree of FIG. 21 without slowing down (in the right lateral direction of FIG. 21).
FIG. 22B shows an example in which the speed is slightly reduced (in the lower right lateral middle direction of FIG. 21) from the initial state in the state transition tree of FIG.
FIG. 22C shows an example in the state transition tree of FIG. 21 in which the speed is reduced (in the downward right lateral direction of FIG. 21) from the initial state.
As can be seen by comparing FIGS. 22 (a) to 22 (c), the larger the deceleration speed, the lower the possibility of collision, but if the deceleration speed is too fast, the sensor detection rate is low (reliability is low). It is the calculation of the collision rate (inside).

By forming a state transition tree for behavior in this way, it is possible to evaluate the collision rate by the algorithm in these networks.

Next, the algorithm in the state transition tree for behavior will be described.
FIG. 23 is a diagram illustrating algorithm in the state transition tree for the behavior of FIG.
In the state transition tree for an action, not all subtrees need to have the same algorithm.
For example, as shown in the alternate long and short dash line box p in FIG. 23, in the state transition tree for the action of FIG. When it is detected, the maximum brake is set regardless of the state transition tree for the action shown in FIG. 21, and the collision damage is reduced.

<Continuous treatment of collision rate by algorithm α>
The continuous treatment of the collision rate by the algorithm α will be described.
FIG. 24A is a diagram illustrating the continuous treatment of the collision rate by the algorithm α.
In FIG. 24 (a) _{, the state is changed to S i} by adding a disturbance caused by various external factors to _{the action A i} selected by the algorithm α to transition to the state S _{i in the state S init.} It does not always transition, and represents a model in which state transitions have a continuous probability distribution.

In FIG. 24 (a), the probability p that a transition to a state _{S i} with the action _{A i} (s | _A) has a normal distribution with standard deviation s _sigma. As a result, collision rate generated in the state _{S i} p (C | s) and state _S discomfort index occurring at _i p (U | s) also have a continuous distribution characteristic shown in FIG. 24 (a). In this way, by assuming a probabilistic model that assumes disturbance, it is possible to more accurately handle probabilistic events that are actually observed.

On the other hand, FIG. 24 (b) shows a model when the disturbance is not considered. In this case, since there is no disturbance, the distribution of p (s | A) does not have a spread and has a discontinuous characteristic. Become. Similarly, the collision rate p (C | s) and the discomfort rate p (U | s) also have discontinuous characteristics, and it becomes impossible to properly handle stochastic events actually observed.

The method of handling the probability in the model of FIG. 24A will be described below. In FIG. 24 (a) and each of the following equations, A (A _i ) is an action, S is a state ( _Sinit is an initial state), C is a collision, U is an unpleasant state, D is an obstacle detection by a distance detection unit, and D. ￣ indicates no obstacle detected by the distance detection unit, E indicates the presence of an obstacle, E￣ indicates no obstacle, and d ^ indicates detection information (detection presence / absence, detection distance, etc.) by the distance detection unit. Note that A ₁ represents a cruising operation.

The algorithm α surrounded by the broken line in FIG. 24 (a) is _{an action A i} when the total collision rate is the closest value to the _{predetermined collision rate P c of α = {D → A i} , D￣ → A _1}. Ask for.
The positive detection rate of the distance detection unit: p (D | E), the false detection rate of the distance detection unit: p (D | E￣), and the detection information d ^ of the distance detection unit are input to the algorithm α.

The collision rate p (C | A) when the action A is selected is expressed by the equation (8), _{and the action AD when the obstacle is detected and the action AD￣} when the obstacle is not detected are given by the equation (9). Indicated by.
Then, the total collision rate p (C | E) is represented by the formula (10) from the formula (8) and the formula (9).

As shown in the equation (8), the collision rate p (C | A) when the action A is selected is the probability p (s | A) of which state the action A transitions to and the collision rate in the state s. It is expressed as information having a continuous distribution, multiplying them, and integrating them in the state s. In the embodiment of the present application, the collision rate is continuously treated based on the model on the premise that a disturbance is applied to the action A in this way.

[Drafting an action plan]
The action plan generated by integrating the detection error and the detection rate of the sensor will be described.
FIG. 25 is an image diagram showing an outline of the action plan, in which the horizontal axis represents the actual obstacle distance d and the vertical axis represents the observed obstacle distance d ^.
As shown in FIG. 25, the action plan includes "sudden braking" used when the observed obstacle distance d ^ is short and "preliminary braking" used when the observed obstacle distance d ^ is shorter.

<Action plan generated by integrating sensor detection error and detection rate>
FIG. 26 shows the action plan of FIG. 25 and the result of the cruising operation executed when no obstacle is detected as obstacle detection D and obstacle non-detection D￣, obstacle existence E and obstacle-free E￣. It is an image diagram represented by a matrix composed of. In FIG. 26, a star mark indicates that an obstacle is overlooked even though it is at a short distance, and a star mark indicates that an obstacle is erroneously detected at a short distance even though there is no obstacle. When an obstacle is detected by the sensor, a braking operation is executed, and when an obstacle is not detected, a cruising operation is executed.

-In the case of obstacle presence E and obstacle detection D, it is "positive detection", and it is an action plan in which sudden braking or preliminary braking is performed according to the observed obstacle distance.
If the observed obstacle distance d ^ is farther than the actual obstacle distance d, it is a preliminary braking that may cause a collision (it may be relieved in a later action plan).
When the observed obstacle distance d ^ is closer than the actual obstacle distance d, the braking is too early.
When the observed obstacle distance d ^ is approximately equal to the actual obstacle distance d, the correct sudden braking or the correct preliminary braking is performed according to the actual obstacle distance d.

-In the case of obstacle existence E and obstacle non-detection D￣ In this case, since the obstacle is not recognized, the cruising operation is executed as an action plan. This is a "miss" and can lead to a collision.
When the actual obstacle distance d is a short distance, it is a cruising with a possibility of collision immediately, and it becomes necessary to quickly execute the collision avoidance action.
If the actual obstacle distance d is long, it is a cruise that may collide later (may be rescued in a later action plan).

-In the case of obstacle detection D with no obstacle E￣ Although it is a "false positive", the action plan executes the braking operation because the obstacle is erroneously recognized. When the observed obstacle distance d ^ is short, useless sudden braking is performed, and when the observed obstacle distance d ^ is long, useless preliminary braking is performed.

-In the case of no obstacle E￣ and no obstacle detection D￣, it is "correct / non-detection". In this case, the action plan executes the cruising operation because the obstacle is not recognized. This is the correct cruise.

In this way, in the case of obstacle presence E and obstacle non-detection D￣, overlooking the obstacle even though it is at a short distance (see the star mark) leads to a decrease in safety.
Further, in the case of obstacle detection D with no obstacle E￣, false detection that there is an obstacle in a short distance even though there is no obstacle (see the star mark) leads to a decrease in comfort and security.
This embodiment suppresses the above-mentioned "missing" and "false positive".

<AND detection>
FIG. 27 is an image diagram illustrating AND detection in the action plan using the redundancy of the two sensors. FIG. 27 shows the results of executing the action plan of FIG. 25 using AND detection, with obstacle detection D ₁ and obstacle non-detection D￣ ₁ based on one sensor, and obstacles based on the other sensor. It is represented by a matrix consisting of detection D ₂ and obstacle non- _{detection D ￣ 2} , obstacle presence E and obstacle-free E ￣. In the case of AND detection, the braking operation is adopted only when an obstacle is detected by both sensors at the same time, that is, when D ₁ and D ₂ occur at the same time, and the cruising operation is adopted in other cases. In FIG. 27, a star mark indicates that an obstacle is overlooked even though it is at a short distance, and a star mark indicates that an obstacle is erroneously detected at a short distance even though there is no obstacle.
The AND detection shown in FIG. 27 corresponds to the example of detecting the two sensors by AND described in FIG. 18 (a).

· Obstacle obstacle exists E detection D ₁ and the obstacle is not detected D¯ "two contradictory state of the sensor" such as the case of ₂ obstacle is determined not to exist. In AND detection, even if an obstacle is detected (D ₁ ) by one of the sensors, it is determined that the sensor information is inaccurate due to a failure of one of the two sensors. In this case, the braking may not be correct and an accident may occur. Therefore, the cruising operation is adopted.
When the actual obstacle distance d is a short distance, it is a cruising with a possibility of collision immediately, and it becomes necessary to quickly execute the collision avoidance action.
If the actual obstacle distance d is long, the cruise will be a possibility of collision later.

· If the obstacle presence E obstacle non-detection D - ₁ and the obstacle non-detection D - ₂ In this case, cruise operation is employed. This is a "missing / erroneous cruising" such as judging that there are no obstacles, which may lead to a collision accident.
When the actual obstacle distance d is a short distance, it is a cruising with a possibility of collision immediately, and it becomes necessary to quickly execute the collision avoidance action.
If the actual obstacle distance d is long, the cruise will be a possibility of collision later.

- the obstacle presence E such case an obstacle of the obstacle non-detection D - ₁ and the obstacle detection D ₂ is determined that there is a "conflict state of the sensor." In AND detection, even if an obstacle is detected (D ₂ ) by one of the sensors, it is determined that the sensor information is inaccurate because one of the two sensors has a failure or the like. In this case, the braking may not be correct and an accident may occur. Therefore, the cruising operation is adopted.
When the actual obstacle distance d is a short distance, it is a cruising with a possibility of collision immediately, and it becomes necessary to quickly execute the collision avoidance action.
If the actual obstacle distance d is long, the cruise will be a possibility of collision later.

-In the case of obstacle detection D ₁ and obstacle detection D ₂ with no obstacle E￣ In this case, the braking operation is adopted. This is "false positive / false braking", and when the observed obstacle distance d ^ is short, it becomes useless sudden braking, and when the observed obstacle distance d ^ is far, it becomes useless preliminary braking. It becomes.

As described above, in the AND detection, the cruising operation is adopted except in the case of the _{obstacle detection D 1} and the obstacle detection D _2. This cruising operation is a cruising with a possibility of collision in the case of the presence of an obstacle E. On the other hand, in the case of E￣ without obstacles, the cruising operation adopted other than the cases of _{obstacle detection D 1} and obstacle detection D _{2 is correct cruising.}

<OR detection>
FIG. 28 is an image diagram illustrating OR detection in the action plan using the redundancy of the two sensors. FIG. 28 shows the results of executing the action plan of FIG. 25 using OR detection, with obstacle detection D ₁ and obstacle non-detection D￣ ₁ based on one sensor, and obstacles based on the other sensor. It is represented by a matrix consisting of detection D ₂ and obstacle non- _{detection D ￣ 2} , obstacle presence E and obstacle-free E ￣. In the case of OR detection, the braking operation is adopted when an obstacle is detected by either or both sensors, that is, when D ₁ and / or D ₂ occur, and the cruising operation is adopted in other cases. In FIG. 28, a star mark indicates that an obstacle is overlooked even though it is at a short distance, and a star mark indicates that an obstacle is erroneously detected at a short distance even though there is no obstacle.
The OR detection shown in FIG. 28 corresponds to the example of detecting two sensors by OR described in FIG. 18 (b).

· If the obstacle presence E obstacle non-detection D - ₁ and the obstacle non-detection D - ₂ In this case, cruise operation is employed. This is a "missing / erroneous cruising" such as judging that there are no obstacles, which may lead to a collision accident.
When the actual obstacle distance d is a short distance, it is a cruising with a possibility of collision immediately, and it becomes necessary to quickly execute the collision avoidance action.
If the actual obstacle distance d is long, the cruise will be a possibility of collision later.

-In the case of obstacle detection D ₁ and obstacle detection D ₂ with no obstacle E￣ In this case, the braking operation is adopted. This is "false positive / false braking", and when the observed obstacle distance d ^ is short, it becomes useless sudden braking, and when the observed obstacle distance d ^ is far, it becomes useless preliminary braking. It becomes.

• If the obstacle without E¯ of obstacle detection D ₁ and the obstacle non-detection D - ₂ In this case, the braking operation is employed. When the observed obstacle distance d ^ is short, useless sudden braking is performed, and when the observed obstacle distance d ^ is long, useless preliminary braking is performed.

· If the obstacle without E¯ obstacle non-detection D - ₁ and the obstacle detection D ₂ In this case, the braking operation is employed. When the observed obstacle distance d ^ is short, useless sudden braking is performed, and when the observed obstacle distance d ^ is long, useless preliminary braking is performed.

As described above, in the OR detection, the cruising operation is adopted only when the _{obstacle non-detection D￣ 1} and the obstacle non-detection D￣ _{2 occur at the same time.} If this cruising operation is performed in the state of presence E with obstacles, it will be a “missed / erroneous cruising” cruising, and if it is performed in the state of E￣ without obstacles, it will be a correct cruising.

Comparing the above-mentioned AND detection (see FIG. 27) and OR detection (see FIG. 28), the AND detection emphasizes the sensor detection state and regards the sensor failure as a factor leading to the possibility of collision. Therefore, it is possible to prevent the sensor failure from being overlooked and delaying the determination of the possibility of collision, and it can be expected that the reliability will be further improved. In the case of AND detection, if a sensor inconsistency occurs due to a sensor malfunction, cruising operation is selected. As a result, comfort is not lost in the case of E￣ without obstacles. However, in the case of the presence of an obstacle E, the safety may be impaired.

In the case of OR detection, if a sensor inconsistency occurs due to a sensor malfunction, braking operation is selected. This results in loss of comfort in the case of unobstructed E￣. However, in the case of the presence of an obstacle E, safety is ensured.

<HALF AND detection>
FIG. 29 is an image diagram illustrating HALF AND detection in the action plan using the redundancy of the two sensors. The same parts as those of AND detection in FIG. 27 are designated by the same reference numerals, and the description of overlapping parts will be omitted.
The HALF AND detection shown in FIG. 29 shows an example in which braking is performed according to the detection states of the two plurality of sensors described in FIG. 18 (c).
In the AND detection of FIG. 27, the "sensor inconsistency state" is recognized in _{the case of obstacle detection D 1} and obstacle non- _{detection D ￣ 2 and} the case of obstacle non- _{detection D ￣ 1} and obstacle detection D _2. Was. In the HALF AND detection of FIG. 29, the “sensor inconsistency state” is recognized as a inconsistency state only when the above conditions are satisfied and the observed obstacle distance d ^ is equal to or greater than a predetermined threshold value.

As shown in FIG. 29, in HALF AND detection, obstacle detection D ₁ and obstacle non- _{detection D ￣ 2} and obstacle non- _{detection D ￣ 1} and obstacle detection D ₂ were observed. When the obstacle distance d ^ is equal to or greater than a predetermined threshold value, it is regarded as a “sensor inconsistency state”, it is determined that there is no obstacle, and the cruising operation is adopted. The result in this case is the same as in the case of AND detection. That is, in the presence of an obstacle E, if the actual obstacle distance d is a short distance, the cruising may be a collision immediately, and if the actual obstacle distance d is long, the cruising may be a collision later. It will be a correct cruise in E￣ without obstacles.

As shown in FIG. 29, in the HALF AND detection, the obstacles observed in the case of _{obstacle detection D 1} and obstacle non- _{detection D ￣ 2 and in} the case of obstacle non- _{detection D ￣ 1} and obstacle detection D ₂ When the object distance d ^ is smaller than the predetermined threshold value, it is determined that an obstacle exists and the preliminary braking operation is adopted. In the presence of an obstacle E, when the actual obstacle distance d is a short distance, a preliminary braking operation with a possibility of collision is performed, and when the actual obstacle distance d is long, a correct preliminary braking operation is performed. In E￣ without obstacles, it becomes useless preliminary braking.

Thus, in HALF AND detection, in the case of obstacle detection D ₁ and obstacle non-detection D￣ ₂ , and in the case of obstacle non- _{detection D ￣ 1} and obstacle detection D ₂ , the observed obstacle distance d ^ A cruising operation or a preliminary braking operation is adopted based on the magnitude relationship between and a predetermined threshold value. As a result, when a sensor inconsistency occurs due to a sensor malfunction, comfort is ensured when the obstacle distance d ^ observed in the case of E￣ without obstacles is equal to or greater than a predetermined threshold value, and the presence of obstacles E Safety is ensured when the obstacle distance d ^ observed in the case is smaller than the predetermined threshold value.

Further, the OR detection of FIG. 28, in the no obstacle E, in the case of the obstacle detection D ₁ and the obstacle non-detection D - _2, in the case of the obstacle non-detection D - ₁ and the obstacle detection D ₂ As a result of adopting "sudden braking" or "preliminary braking" according to the observed obstacle distance d ^, "useless sudden braking" and "useless preliminary braking" were obtained. In HALF AND detection, as a result of adopting "preliminary braking" or "cruising" according to the obstacle distance d ^ observed in the above case, "useless preliminary braking" or "correct cruising" was obtained. This has the effect of reducing the possibility of unnecessary sudden braking and unnecessary preliminary braking increasing as compared with OR detection so as not to impair comfort as much as possible.

In the AND detection shown in FIG. 27, there are 6 ★ marks indicating the case where an obstacle is overlooked even though it is at a short distance, and a ☆ mark indicating a case where an obstacle is erroneously detected at a short distance even though there is no obstacle. There were two, for a total of eight. In the OR detection shown in FIG. 28, there were two ★ marks and six ☆ marks, which were also a total of eight. In the HALF AND detection shown in FIG. 29, there were 4 ★ marks and 2 ☆ marks, for a total of 6 marks.

From this result, it can be said that HALF AND detection is superior in comfort as compared with AND detection, and is superior in comfort as compared with OR detection, but inferior in safety. On the other hand, when compared from a comprehensive point of view, that is, the total number of ★ and ☆ marks, which is a problem, it can be said that HALF AND detection provides improved results for AND detection and OR detection.

The action plan using AND detection, OR detection, and HALF AND detection has been described above. The details of how to obtain the collision rate in each state during traveling will be described below.

[Event rate]
Next, how to obtain the event occurrence rate will be described. The event occurrence rate is the collision rate when "collision" is applied to the event. The event occurrence rate is a generalized concept of the collision rate.
FIG. 30 is a state transition diagram of an event in which a collision occurs.
In FIG. 30, _Sc represents the current state, d ^ represents the observed value, α represents the action, _Sn represents the next state, and C represents an event (collision). State transition diagram of FIG. 30 will observe the current state S _c Oite observations d ^ transitions result of taking action alpha (deceleration) to the next state S _n contrast, that the resulting collision occurs Represents. Based on this state transition model, the probability a collision from the state _{S c} occurs p (C | _{S c)} is expressed by Equation (11).

<How to find the collision rate>
FIG. 31 is a diagram illustrating how to obtain the collision rate under the condition that the actual location of the obstacle is known. FIG. 31 (a) shows the locus of the position and speed when decelerating from the current position and speed and trying to stop at the target stop position set in front of the obstacle, and the horizontal axis is the position and the vertical axis is the speed. Take. FIG. 31 (a) corresponds to FIG. 10A.
FIG. 31 (b) shows the collision rate at zero speed, and FIG. 31 (c) shows the error distribution of target detection by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120). 31 (d) shows the detection rate of target detection by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120). The horizontal axis of FIGS. 31 (b)-(d) shows the position, and the vertical axis shows the occurrence rate. Figure 31 (a) - In _{(d), p (ε α} ) is the probability of uncertainty epsilon _alpha action plan, d ^ ₁ value if it is a value (observed indicating whether obstacle is observed 1 If not observed, the value is 0), d ^ ₂ is the observed obstacle distance, x _c is the current position, x _o is the position of the obstacle, v _c is the speed at the current position, d _M Represents the stop distance margin set between the observed obstacle distance d ^ ₂ and the target position at which the vehicle is stopped.

FIG. 32 is a state transition diagram showing the process of collision generation in FIG. 31.
In Figure 32, action (reduction): alpha, current state: _{S c,} observed value: d ^, the next state: _{S n,} event (collision): a C. Figure 31 (a), expressed the current state _{S c} as defined by the current velocity _{v c} and the position _{x c} and _{(v c, x} c), the observed value d ^ indicates the case of the detection / non-detection d ^ _{1 = {1,0}, defined by d ^ 2} = x ₀ −x _c indicating the distance to the target (d ^ ₁ , d ^ ₂ ), and the next state _Sn is the next state It is defined by the velocity v _n and the position x _{n of (vn n} , x _n ).

The position / velocity locus (thick solid line) shown in FIG. 31 (a) has an uncertainty distribution p (ε _α ).
Here, ε _α represents the uncertainty for the next state transition in the action plan.
The current state _{S c} is the velocity _{v c} and the position _{x c: S c = (v} c, x c)
The observed value d ^ is detected / undetected: d ^ ₁ = {1,0}, distance to the target: d ^ ₂ = x ₀ −x _c
The stopping distance margin is d _M
Deceleration a current state _{S c = (v c, x} c) deceleration which is executed in an attempt to stop at the target stop position _d ^ 2 -d _M In, represented by the formula (12). The velocity v _n and the position x _n representing the next state _Sn are represented by the equations (13) and (14). ΔT in equations (13) and (14) represents the time difference between the current state and the next state. The next state S _n is a state that would transition when deceleration ΔT time deceleration a from the current state S _c.

_{The collision rate at v c} = 0 shown in FIG. 31 (b) is represented by a characteristic having a cumulative distribution of a normal distribution having a standard deviation of σ _d. That is, the collision rate p (C | v) at a velocity of 0
_c = 0, x _c ) is expressed by the equation (15).

The error distribution of target detection by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) shown in FIG. 31 (c) is p (d ^ ₂ | x _c , v _c ). Shown.
Here, d ^ ₂ is the detection distance observed in that state, and has a distribution that depends on _{x c.}

As shown in FIG. 31 (d), the detection rate of target detection by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) is p (d ^ ₁ = 1 | x _c , v). _c ), which changes according to the distance to detection.

v collision rate in _{c = 0 p (C | v} c = 0, x c) is represented by the formula (15). In equation (15), σ _d is the standard deviation of the normal distribution of the collision rate, and erf is a Gaussian error function. Algorithm probability p (epsilon _alpha) of (deceleration) uncertainty of alpha epsilon _alpha has a normal distribution, represented by the formula (16). In other words, p (ε _α ) is the probability taken by the _{uncertainty ε α,} which represents the difference between the deceleration a command and the deceleration actually executed by the vehicle. In equation (16), σ _α is the standard deviation of the normal distribution that represents the uncertainty. Further, in the state (x _c , v _c ), the probability (error distribution) p (error distribution) of detecting the _{target at the position of the distance d ^ 2} by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120). d ^ ₂ | x _c , v _c ) is represented by the equation (17), and is an object by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) _{in the state (x c} , v _c). The detection rate p (d ^ ₁ = 1 | x _c , v _c ) for detecting the target is expressed by the equation (18). In the formula (18), P _max and P _min are the maximum detection rate and the minimum detection rate of the target detection by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120), respectively. d _s and d _e is the parameter representing the distance detection unit (detection device DD, the vehicle sensor 60, and the automatic driving control unit 120) position depends on the characteristics of target detection by the distance detection unit (detection during this period in the region The detection rate of target detection by the device DD, the vehicle sensor 60, and the automatic driving control unit 120) changes from P _{min to P max} (see FIG. 31 (d)).

<Calculation of collision rate p (C | S)>
Next, the calculation of the collision rate p (C | S) for each position / velocity will be described.
33 to 35 are views in which the state is represented by a grid in the two-dimensional space of the position and the velocity in order to calculate the collision rate p (C | S) for each position and velocity.

1. 1. Preparing a Grid FIG. 33 is a diagram showing a grid.
The grid shown in FIG. 33 is prepared. In order to obtain the collision rate p (C | S) for each position / velocity, the state is divided into grids (see FIG. 33), and the calculation is performed in order from the known position.

Grid index: G _x , G _v
Grid size: G _xsize , G _vsize
State (position, velocity): x, v
Maximum position and velocity: x _max , v _max
Then, the index → state value is represented by the formulas (19) and (20), and the state value → index is represented by the formulas (21) and (22) (see FIG. 33).

2. Giving conditions at the edges of the grid of the collision rate map FIG. 34 is a diagram illustrating that conditions are given at the edges of the grid of the collision rate map.
The collision rate is set to 1 when the target is exceeded.
p (G _xmax , G _v ) = 1

The collision rate p (G _x , 0) at a speed of 0 is the cumulative distribution of the target detection distribution by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120), and is expressed in equation (15). Based on the formula (23).

3. 3. The collision rate is obtained in order from the edge of the grid of the collision rate map. FIG. 35 is a diagram for explaining that the collision rate is obtained in order from the edge of the grid of the collision rate map.
As shown by arrows q and r in FIG. 35, the collision rate is obtained in order from the edge of the grid of the collision rate map. Reference numeral s in FIG. 35 indicates an image of a calculation for obtaining the collision rate in a certain state.
Here, the collision rate on the grid is expressed _{as p (C | S) = pc} (G _x , G _v).
Collision rate in current state _{S c p (C | S c} ) is represented by the formula (24), the collision rate _p c on the grid _{(G xc, G vc)} is represented by the formula (25).

以上、各位置・速度に対するコリジョン率ｐ（Ｃ｜Ｓ）の算出について説明した。

The calculation of the collision rate p (C | S) for each position / velocity has been described above.

<Method of approximating the collision rate from the following states>
Next, a method of approximating the collision rate from the following states will be described.
36 to 38 are diagrams illustrating a method of approximating the collision rate from the following states.

1. 1. Obtaining the current speed and position from the position of the grid FIG. 36 is a diagram corresponding to FIG. 31 (a).
From FIG. 36, the current velocity v _c and the position x _c are calculated according to the equations (26) and (27).

2. Set the parameters of d ^ ₁ , d ^ ₂ , ε _α

3. 3. Finding the required deceleration The required deceleration a is calculated according to equation (28).

4. Finding the time to be in the next state FIG. 37 is a diagram illustrating a method of finding the time to be in the next state. x _step represents the axial grid spacing representing the position, and v _step represents the axial grid spacing representing the velocity. Further, ΔT _x represents the time until the vehicle reaches the grid line x = x _c + x _{step , and ΔT v} represents the time until the vehicle reaches the grid line v = v _c −v _step . FIG. 37 (a) _{shows the case of ΔT x} <ΔT _v , and FIG. 37 (b) shows the case of ΔT _x ≧ ΔT _v .
First, as shown in FIGS. 37 (a) and 37 (b), _{ΔT x} , which is the intersection with _{x c} + x _{step, and ΔT v,} which is the intersection with v _{c −} v _step , are calculated according to equations (29) and (30). do.

Next, of the calculated ΔT _x and ΔT _v , the smaller one is adopted according to the equation (31).
ΔT = min (ΔT _v , ΔT _x )… (31)

5. Find the next state Find the next states v _n , x _n by equations (32) and (33).
v _n = v _c + aΔT… (32)
x _n = x _c + v _n ΔT + 1/2 · aΔT ² … (33)

6. Obtaining the collision rate of the next state from the grid points in the vicinity of the next state FIG. 38 is a diagram illustrating a method of obtaining the collision rate of the next state from the grid points in the vicinity of the next state (a). ) Indicates the case of ΔTx <ΔTv, and (b) indicates the case of ΔTx ≧ ΔTv.
As shown in FIGS. 38 (a) and 38 (b), _{grid points (G xc} , G _vc -1 _{) in the vicinity of the following states (g xn} , g _vn ) (see the ● mark in FIGS. 38 (a) and 38 (b)) _), calculated _{(G xc + 1, G vc} -1), (G xc + 1, the collision rate at _{G vc),} according to equation (34), the collision rate of the next state _{p c (g xn, g vn} ) do. Here, _{m x} is the current state _{(G xc,} G _vc) next state from _{(g xn,} g _vn) is a value of the distance normalized by the _{x step} up, thus a value between 0 and 1 Take. m _v is a value obtained by normalizing the velocity difference from the next state (g _xn , g _vn ) to (G _xc + 1, G _vc -1) with v _step , and therefore takes a value between 0 and 1. Since the distance between the grids is 1, the adjacent state can be calculated by +1 or -1.

7. _{d ^ 1, d ^ 2,} ε collision rate of the next state calculated accumulates while sweeping the parameters _{α p c (g xn, g} vn) of the original, by the equation (35), d By calculating the sum while sweeping the parameters of ^ ₁ , d ^ ₂ , and ε _{α , the collision rate p c} (G _xx , G _vc ) of the current state is obtained.

Please note the following. _{P c (g xn, g vn} ) of formula (35) is given by equation _{(34), m x, m} v of formula (34) is determined by the following conditions _{(g xn,} g _vn), the next state (G _xn , g _vn ) is given by equations (32) and (33), equations (32) and (33) are functions of deceleration a and ΔT, and ΔT is ΔT _v , ΔT according to equation (31). _Determined from x, ΔT _v and ΔT _x are given by the function of deceleration a shown in equations (29) and (30). The deceleration a is then given by Eq. (28) by a function containing the _{uncertainty ε α of the deceleration a.} That is, the following states (g _xn , g _vn ) are obtained by adding the _{uncertainty ε α to the deceleration a.}

That is, _{p c (g xn, g vn} ) the values uncertainty epsilon _alpha is reflected deceleration a in the formula (35). The subject of the sum in equation (35) includes a _{p c (g xn, g vn} ) term deceleration a uncertainty epsilon _alpha probability P (epsilon _alpha) is multiplied, for uncertainty epsilon _alpha Includes sweep. Thus, collision rate _{p c (G xc, G vc} ) of the current state of Formula (35) gives the a value uncertainty epsilon _alpha is reflected deceleration a. This makes it possible to handle the collision rate continuously.

Further, the target of the sum of the equation (35) is multiplied by _{the detection rate p (d ^ 1} ) of the target detection by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120). Please be careful. _{The detection rate p (d ^ 1} ) of target detection by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) is represented by the equation (18), and the characteristics shown in FIG. 31 (d) are exhibited. Have. _{In the region where the detection rate p (d ^ 1} ) of target detection by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) is low, that is, in the region where the distance from the target is long, the equation (35) collision rate _{p c (G xc, G vc} ) by is evaluated lower. In the region where the collision rate is evaluated low on the collision rate map, as explained with reference to FIG. 8, the action plan generation unit 200 can generate an action plan such that the vehicle starts decelerating within a range that does not affect the riding comfort. It becomes.

<How to find the approximate collision rate map>
The collision rate map set in step S12 of the flow according to the embodiment of the present application is approximately calculated by the equations (16) to (18), (23), (26) to (35) assuming the situation shown in FIG. It was done. That is, the collision rate map is calculated assuming that there is an obstacle in the detection distance detected by the sensor configuration that currently detects the obstacle. This collision rate map is hereinafter referred to as an "approximate collision rate map".

The term "approximate collision rate map" used here is different from the "collision rate map" in which the true position of the target is known as described above, and is different from the distance detection unit (detection) as shown in FIG. Since it is calculated on the assumption that there is an obstacle _{at the detection distance d ^ 2} detected by the device DD, the vehicle sensor 60, and the automatic driving control unit 120), it is used for convenience of explanation.

In FIG. 36, d ^ ₂ and d _M of the equation (28) are a plurality of detection devices DD among the detection information output by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120). It corresponds to the detection distance, which is the distance detected by the combination of sensors in which obstacles are detected (sensor configuration), and the offset amount determined by the command value offset amount calculation unit 230, respectively.

The collision rate at each position where the velocity v = 0, which is the end of the approximate collision rate map used in the calculation, is given by Eq. (23). However, the value of the detection distance is substituted for _{the true value x o} of the position of the target. Further, it is assumed that the collision rate at the _{position x c} + d ^ ₂ , which is the other end of the approximate collision rate map, and v ≠ 0 is 1.

Based on the collision rate at each position at the velocity v = 0 and the collision rate at the position x _c + d ^ ₂ and v ≠ 0, the collision rate at the point on the approximate collision rate map is approximately obtained according to Eq. (34). Be done.

P (d ^ ₁ ) of equation (35) is given by applying equation (18) at each grid point on the approximate collision rate map. In equation (18), x _c is given the position x given in equation (19).

P (d ^ ₂ ) in equation (35) is calculated using equation (17). However, substitutes detection distance to the true value x _o of the target position. P (d ^ ₂ ) in equation (35) corresponds to "probability distribution 1003 based on fusion accuracy reliability". As described above, the probability distribution 1003 based on the fusion accuracy reliability is a distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) in a situation where the true value of the distance to the target is known. ) Can be used to obtain measured values based on the results of the process of measuring the characteristics of target detection. The width of the probability distribution 1003 based on the fusion accuracy reliability depends on the detection distance, and the larger the detection distance, the wider the normal distribution. _{Therefore, the parameter σ d} of the normal distribution characteristic expressed by the equation (17) is obtained in advance as a function of the detection distance of the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) based on the measured value. Then, set the value according to the detection distance. It should be noted that this σ is also used in the equation (23) that gives the collision rate at each position where the velocity v = 0.
Use the value of _d.

_{P (ε α} ) of equation (35) is given by equation (16). _{The parameter σ α} of the normal distribution characteristic expressed by Eq. (16) can be obtained in advance by prior measurement. The value calculated by Eq. (16) using this measured value is used as the value of p (ε _α ).

_{P c (g xn, g vn} ) of formula (35) is given by equation (34).

Offset amount command value used in calculating the "approximate collision rate map" (i.e. d _M) has a predetermined target collision rate, a command value achieved probability density distribution 1001, from the probability distribution 1003 Metropolitan based on fusion accuracy reliability degrees It was what I wanted. The command value achievement probability density distribution 1001 is measured in advance through a process of measuring the stop position when a certain deceleration is commanded to the vehicle 1, and can be regarded as a function of deceleration. The probability distribution 1003 based on the fusion accuracy reliability is a characteristic to be measured in advance through a process of testing what the distribution of the error of the detection distance will be in a situation where the true value of the distance is known, and is a distance detection unit (detection). It can be regarded as a function of detection information (sensor configuration, detection time, and detection distance) output by the device DD, the vehicle sensor 60, and the automatic driving control unit 120).

From the above explanation, the "approximate collision rate map" is the detection information (sensor configuration, detection time, and detection distance) output by the deceleration and distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120). You can see that it can be regarded as a function.
<Method of acquiring an approximate collision rate map during traveling in the embodiment>

In the embodiment, the "approximate collision rate map" assumes a plurality of situations in advance, creates a plurality of "approximate collision rate maps", stores them in the collision rate map storage unit 1010, and stores the "approximate collision rate map" in the collision rate map storage unit 1010 while the own vehicle 1 is traveling. The collision rate map setting unit 250 selects the one that is close to the current situation.

As described above, the "approximate collision rate map" is the deceleration, detection information (sensor configuration, detection time, and detection distance) output by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120). It can be regarded as a function. Therefore, a table (approximate collision rate map table) (not shown) that uses deceleration, sensor configuration, detection time, and detection distance as input parameters and outputs an "approximate collision rate map" is created in advance and is stored in the collision rate map storage unit. It is stored in 1010. Then, while the own vehicle 1 is traveling, the collision rate map setting unit 250 determines the maximum deceleration currently set by the action plan generation unit 200, and the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit). The sensor configuration indicating the combination of sensors currently detecting obstacles obtained from 120), the detection time continuously detected by the sensor configuration, and the detection distance are approximated stored in the collision rate map storage unit 1010. Input to the collision rate map table and determine the "approximate collision rate map" output from the table as the "approximate collision rate map" currently used.

By using this approximate collision rate map table, it is possible to set the "approximate collision rate map" at high speed with less consumption of calculation resources.

This approximate collision rate map table can store a more accurate "approximate collision rate map" as the amount of information of the input parameters increases (the smaller the quantization width), but the capacity increases. The amount of information input should be determined by the trade-off between accuracy and capacity.

Furthermore, the table capacity can be reduced by not storing the collision rates at all the grid points on the "approximate collision rate map". For example, the collision rate in a region where the collision rate is substantially higher or lower than the target collision rate is treated as a fixed value, and the data format can be determined so as not to be stored in the collision rate map storage unit 1010.

Unlike the above-described embodiment in which the collision rate map table is stored in the collision rate map storage unit 1010 and the collision rate map table is selected while the own vehicle 1 is running, the collision rate map setting unit 250 is used. , The maximum deceleration currently set while driving, the current sensor configuration obtained from the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120), the detection time, and the detection distance in advance. It may be calculated in real time according to various characteristics of the own vehicle 1 that have been measured. According to this embodiment, the collision rate map storage unit 1010 can be omitted. In this embodiment, the offset amount determined by the command value offset amount calculation unit 230 described in relation to STEP 4 described above is used, and the equation (16) is described as a method of obtaining the approximate collision rate map described above. )-(18), (23), (26)-(35), the approximate collision rate map can be calculated.

Further, in the case of the embodiment in which the "approximate collision rate map" is calculated in real time, the calculated "approximate collision rate map" is stored in the collision rate map storage unit 1010 as an element of the approximate collision rate map table, and is similar later. You may read and use the approximate collision rate map that is stored when you encounter a situation. Further, the infrequently used approximate collision rate map may be deleted from the approximate collision rate map table of the collision rate map storage unit 1010. According to this embodiment, the capacity required for the collision rate map storage unit 1010 can be reduced.

Further, even in the embodiment in which the "approximate collision rate map" is calculated in real time, the consumption of calculation resources is reduced by not calculating the collision rate at all the grid points on the "approximate collision rate map", and the collision rate map is stored. The capacity for storing the "approximate collision rate map" in the unit 1010 can be reduced.

The details of how to obtain the "approximate collision rate map" in the embodiment have been described above.

The vehicle control device 100 according to the embodiment has an action plan generation unit 200 that generates an action plan for automatic driving of the own vehicle 1, and a vehicle behavior control unit 160 that controls at least the speed of the own vehicle 1 based on the action plan. The action plan generation unit 200 includes a distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) that detects a target and outputs detection information related to the detection of the target. The maximum deceleration of the own vehicle 1 in automatic driving is set, and the action plan generation unit 200 is positioned when the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) detects an obstacle. The collision rate map setting unit 250 for determining a collision rate map, which is a two-dimensional map showing the distribution of the collision rate indicating the possibility of collision between the own vehicle 1 and the obstacle in the two-dimensional space of the speed and the collision rate, is provided. The map is created on the premise of a predetermined target collision rate, the maximum deceleration, and a target stop position set based on the combination of the intellectual information, and the action plan generation unit 200 includes the collision rate map and the collision rate map. A current action plan is generated based on a predetermined target collision rate and the current position and speed of the own vehicle 1.

The conventional technique controls the speed of the vehicle from the position error probability distribution as in Patent Document 1. Further, as in Patent Document 2, the possibility of collision with the own vehicle is determined. Both techniques control the speed of the vehicle from the position error probability distribution and the possibility of collision, and are not associated with the action plan of autonomous driving. Therefore, when the action plan is determined using the sensor, there is a problem (problem 1) that the recognition distance is insufficient and the recognition distance cannot be extended while maintaining high reliability of the sensor. In addition, the level of safety is clarified and the reliability of the sensor (accuracy,
There is a problem (problem 2) that the accuracy of the action plan cannot be quantified.

On the other hand, in the present embodiment, a collision rate map, which is a two-dimensional map showing the distribution of the collision rate indicating the possibility of collision between the own vehicle 1 and the obstacle in the two-dimensional space of position and speed, is introduced. The current action plan is generated from the relationship between the collision rate of the position / speed of the own vehicle 1 on the rate map and the position / speed having the target collision rate on the collision rate map. As a result, it is possible to grasp the relationship between the position and the speed having the target collision rate in the two-dimensional space of the position and the speed, and it is possible to generate an action plan in consideration of the safety factor and the comfort in the automatic driving.

In the collision rate map of the present embodiment, a plurality of grid points are defined in the two-dimensional space, and a collision rate is determined for each grid point. As for the collision rate at each grid point, the own vehicle 1 is the grid point. When the vehicle behavior control unit 160 commands the own vehicle 1 to decelerate with the maximum deceleration as the upper limit so as to decelerate from the position and speed corresponding to the above and stop at the target stop position, the own vehicle 1 causes the obstacle. Represents the probability of colliding with an object.

With this configuration, the action plan generation unit 200 can generate an action plan based on the collision rate premised on the currently set maximum deceleration.

The action plan generation unit 200 according to the present embodiment has a collision rate in which the current position and speed of the own vehicle 1 are lower than the first threshold value in which the collision rate shown on the collision rate map is lower than the predetermined target collision rate. When the vehicle is in the low rate region (see FIG. 14), an action plan that maintains the speed of automatic driving and allows sudden braking is generated as the current action plan, and the collision rate on the collision rate map is the predetermined target. In the collision rate high region (see FIG. 14), which is lower than the collision rate and equal to or higher than the first threshold value and equal to or higher than the second threshold value, an action plan for performing preliminary braking to avoid sudden braking by repeating short-time braking is repeated. As the current action plan.

By doing so, the action plan generation unit 200 is in the two-dimensional space of the collision rate map in the collision rate low region (see FIG. 14) where the collision rate is lower than the first predetermined threshold value lower than the target collision rate. A sudden braking action plan that maintains the set speed of automatic driving and allows sudden braking is generated, and the collision rate is lower than the target collision rate and equal to or higher than the second threshold value, which is equal to or higher than the first predetermined threshold. In the region where the collision rate is high (see FIG. 14), it is possible to generate a preliminary braking action plan that avoids sudden braking by repeating short-time braking. By controlling the automatic driving of the own vehicle 1 in this way, it is possible to achieve both a safety factor and comfort.

The distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) according to the present embodiment includes a plurality of sensors, and the detection information is a sensor that detects the obstacle among the plurality of sensors. In an attempt to stop the vehicle at the target stop position, including a sensor configuration showing the combination of the above, a detection distance detected by the sensor configuration, and a detection time continuously detected by the sensor configuration. Centering on the command value achievement probability density distribution 1001 which is the probability density distribution of the position where the vehicle actually stops when the maximum deceleration is commanded to the own vehicle 1 by the vehicle behavior control unit 160 and the position indicated by the detection distance. The collision rate calculated by convolving the error distribution (probability distribution 1003 based on fusion accuracy reliability) that is distributed and shows the probability distribution of the difference between the true distance to the obstacle and the detection distance is the predetermined target collision. The target stop position is set so as to be equal to the rate, and the command value achievement probability density distribution 1001 is measured in advance by executing the stop operation by the deceleration specified for the own vehicle 1 on the own vehicle 1. , The error distribution (probability distribution 1003 based on fusion accuracy reliability) is estimated for the maximum deceleration based on the vehicle stop characteristic of the own vehicle 1.
The distance detection unit (detection device DD, vehicle) measured in advance using the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) in a situation where the true value of the distance to the target is known. It is estimated from the distance detection characteristics of the sensor 60 and the automatic operation control unit 120) with respect to the sensor configuration, the detection distance, and the detection time.

The error distribution showing the probability distribution of the difference between the detection distance detected by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) and the true distance to the obstacle is the error distribution of the multiple sensors of the distance detection unit. Of these, it can be estimated from the sensor configuration indicating the combination of the sensors that have detected the obstacle, the detection distance detected by the sensor configuration, and the detection time that is continuously detected by the sensor configuration. The overlap between this error distribution and the command value achievement probability density distribution at the maximum deceleration represents the collision rate of the own vehicle 1. By setting the target stop position so that the collision rate becomes a predetermined target collision rate equal to or less than the collision rate expected in manual driving, for example, the case where the automatic driving is driven and the case where the person is driving are compared. In some cases, the collision rate map can be defined so that it is safer when autonomous driving is driving.

Further, by configuring as described above, the action plan generation unit 200 is based on the characteristics actually measured for the own vehicle 1 and the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120). A collision rate map with a defined stop position can be used.

The vehicle control apparatus 100 according to the present embodiment, collision rate map that collision rate map setting unit 250 sets the grid points on the collision rate map in accordance with equation _{(35) (G xc, G} vc) collision rate in _p c ( G _xx , G _vc ) is obtained by Eq. (35).
In the formula (35), p (d ^ ₁ ) is a detection that detects the obstacle by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) at the grid points (G _xx , G _vc). The rate, p (d ^ ₂ ), is the position of the _{obstacle at the distance d ^ 2} by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) at the grid points (G _xx , G _vc). probability of _{detecting, p c (g xn, g} cn) the own vehicle 1 is a grid point _{(G xc, G vc)} vehicle behavior control unit 160 to stop from the target stop position is the deceleration a vehicle _{The collision rate at points (g xn} , g _cn ) on the collision rate map that will transition when commanded to 1 and p (ε _α ) are actually executed by the own vehicle 1 in response to the command of deceleration a. It represents the probability that the _{uncertainty ε α,} which represents the difference from the deceleration, takes.
_{The points (g xn} , g _cn ) on the collision rate map are points _{obtained by adding uncertainty ε α} to the deceleration a, and are on the grid points where the velocity = 0 on the collision rate map. The collision rate is given based on the result of measuring the characteristics of the distance detection unit in advance. A predetermined value is given to the collision rate on the grid points where the position on the collision rate map is the detection distance and the speed is other than 0.
_{p c (g xn, g cn} ) value of is obtained by calculation for obtaining the original to approximate the collision rate at the grid point in the vicinity of the point _{(g xn,} g _cn).
The total calculation of the equation (35) is repeated in the direction in which the speed increases from the grid point corresponding to the speed = 0 and the position = the detection distance and / or in the direction in which the position approaches the own vehicle 1 from the detection distance. The collision rate at the grid points (G _xx , G _{vc) can be obtained.}

The collision rate map obtained in this way is a value that takes into account the _{uncertainty ε α of the deceleration a.} As a result, the collision rate map shows a situation closer to the state transition in actual driving.

The target of the sum of the formula (35) is multiplied by _{the detection rate p (d ^ 1} ) of the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120). _{When the detection distance is long, the detection rate p (d ^ 1} ) of the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) takes a low value. _{In the region where the detection rate p (d ^ 1} ) of the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) is low, that is, in the region where the distance from the target is long, the collision rate is determined by equation (35). It is evaluated low. In the region where the collision rate is evaluated low on the collision rate map, the action plan generation unit 200 can select to start decelerating within a range that does not affect the riding comfort, as explained with reference to FIG. ..

In this embodiment, the probability p (ε _α ) has a normal distribution with a standard deviation determined based on the pre-measured characteristics of the own vehicle 1.

With this configuration, it is possible to obtain a collision rate map based on the uncertainty of the actual deceleration operation of the own vehicle 1.

The action plan generation unit 200 according to the present embodiment is in a situation where the result of the process of measuring the stop position when a certain deceleration is commanded to the own vehicle 1 and the true value of the distance to the target are known. A plurality of collisions obtained from the result of the process of measuring the characteristics of target detection by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) and the predetermined target collision rate. It has a collision rate map storage unit 1010 that stores a rate map. The collision rate map setting unit 250 selects and uses one of the plurality of collision rate maps stored in the collision rate map storage unit 1010 based on the maximum deceleration and the detection information. Determined as a collision rate map.

With this configuration, the collision rate map can be determined without performing the calculation for obtaining the collision rate map during traveling, so that the consumption of calculation resources can be reduced.

The collision rate map setting unit 250 according to the present embodiment is in a situation where the result of the process of measuring the stop position when a certain deceleration is commanded to the own vehicle 1 and the true value of the distance to the target are known. The result of the process of measuring the characteristics of target detection by the distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) in the above, and the own vehicle 1 and the distance detection unit (detection) derived from Calculations are performed in real time during driving based on parameters indicating the characteristics of the device DD, the vehicle sensor 60, and the automatic driving control unit 120), the predetermined target collision rate, the maximum deceleration, and the detection information. The collision rate map may be determined accordingly.

Further, the action plan generation unit 200 according to the present embodiment has a collision rate map storage unit 1010 that stores the collision rate map calculated in real time by the collision rate map setting unit 250 during traveling, and has a collision rate map setting unit. The 250 is a collision rate map storage unit 1010 when a collision rate map corresponding to the predetermined target collision rate, the maximum deceleration, and the detection information is stored in the collision rate map storage unit 1010. It may be determined as a collision rate map using the collision rate map stored in.

With this configuration, the storage resource of the collision rate map storage unit 1010 that stores the collision rate map can be eliminated or reduced.

The distance detection unit (detection device DD, vehicle sensor 60, and automatic driving control unit 120) according to the present embodiment includes a plurality of sensors for detecting an obstacle, and the detection information is obtained from the plurality of sensors. When the obstacle is detected only by a part of the plurality of sensors, including a sensor configuration showing a combination of sensors that have detected the obstacle and a detection distance detected by the sensor configuration. When the detection distance is equal to or greater than a predetermined distance threshold value, the action plan generation unit 200 generates an action plan for executing the cruising operation by assuming that the obstacle does not exist, and generates the plurality of action plans as the current action plan. When the obstacle is detected only by some of the sensors of the above, and the detection distance is smaller than the predetermined distance threshold, the action plan generation unit 200 considers that the obstacle exists. An action plan that implements deemed preliminary braking may be generated as the current action plan.

With this configuration, when a sensor inconsistency occurs due to a sensor malfunction, comfort is ensured when the obstacle distance d ^ observed in the case of E￣ without obstacles is equal to or greater than a predetermined threshold value, and the presence of obstacles E. If the obstacle distance d ^ observed in the above case is smaller than the predetermined threshold value, safety is ensured.

The above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one including all the described configurations. Further, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. .. Further, it is possible to add / delete / replace a part of the configuration of the embodiment with another configuration.

The vehicle control device and the vehicle control method of the present invention are also realized by a program for making the computer function as the vehicle control device and the vehicle control method. The program may be stored on a computer-readable recording medium.

1 Vehicle 20 Finder 30 Radar (sensor) (distance detector)
40 Camera (sensor) (distance detector)
50 Navigation device 55 Communication device 60 Vehicle sensor (sensor) (distance detector)
70 HMI
100 Vehicle control device 110 Target lane determination unit 120 Automatic driving control unit (distance detection unit)
130 Automatic driving mode control unit 140 Recognition unit 141 Own vehicle position recognition unit 142 External world recognition unit 143 Person detection unit 144 AI accelerator 145 Track generation unit 150 Switching control unit 160 Travel control unit (Vehicle behavior control unit)
170 HMI control unit 180 Storage unit 200 Action plan generation unit 210 Target collision rate setting unit 220 Command value achievement probability Density distribution estimation unit 230 Command value offset amount calculation unit (offset amount calculation unit)
240 fusion accuracy reliability calculation unit 250 collision rate map setting unit 300 driving force output device 310 steering device 320 braking device 1000 collision rate map 1001 command value achievement probability density distribution 1002 command value offset amount 1003 probability based on fusion accuracy reliability Distribution 1010 Collision rate map storage unit DD detection device (sensor) (distance detection unit)

Claims (11)

A vehicle control device that controls a vehicle
An action plan generation unit that generates an action plan for automatic driving of the vehicle,
Based on the action plan, at least a vehicle behavior control unit that controls the speed of the vehicle,
It is equipped with a distance detection unit that detects a target and outputs detection information related to the detection of the target.
The action plan generation unit sets the maximum deceleration of the vehicle in automatic driving, and sets the maximum deceleration of the vehicle.
The action plan generation unit is a two-dimensional map showing the distribution of collision rates indicating the possibility of collision between the vehicle and the obstacle in a two-dimensional space of position and speed when the distance detection unit detects an obstacle. It is equipped with a collision rate map setting unit that determines the collision rate map.
The collision rate map is created on the premise of a target stop position set based on a combination of a predetermined target collision rate, the maximum deceleration, and the detection information.
The action plan generation unit is a vehicle control device that generates a current action plan based on the collision rate map, the predetermined target collision rate, and the current position and speed of the vehicle. The collision rate map defines a plurality of grid points in the two-dimensional space.
A collision rate is set for each grid point,
The collision rate at each grid point is determined by the vehicle behavior control unit with the maximum deceleration as the upper limit so that the vehicle decelerates from the position and speed corresponding to the grid point and stops at the target stop position. The vehicle control device according to claim 1, wherein the vehicle represents the probability that the vehicle will collide with the obstacle when a command is given to the vehicle. The action plan generation unit is in the case where the current position and speed of the vehicle are in a collision rate low region where the collision rate shown on the collision rate map is lower than the first threshold value lower than the predetermined target collision rate. Generate an action plan as the current action plan that maintains the speed of automatic driving and allows sudden braking,
In the collision rate high region where the collision rate is lower than the predetermined target collision rate and equal to or higher than the second threshold value which is equal to or higher than the first threshold value on the collision rate map, short-time braking is repeated to perform sudden braking. The vehicle control device according to claim 1 or 2, wherein an action plan for performing preliminary braking to avoid is generated as the current action plan. The distance detection unit includes a plurality of sensors and has a plurality of sensors.
The detection information is continuously detected by the sensor configuration indicating the combination of the sensors that have detected the obstacle among the plurality of sensors, the detection distance detected by the sensor configuration, and the sensor configuration. Including detection time,
The command value achievement probability density distribution, which is the probability density distribution of the position where the vehicle actually stops when the maximum deceleration is commanded to the vehicle by the vehicle behavior control unit in an attempt to stop the vehicle at the target stop position. The collision rate calculated based on the overlap of the error distributions distributed around the position indicated by the detection distance and indicating the probability distribution of the difference between the true distance to the obstacle and the detection distance is the predetermined target collision. The target stop position is set so as to be equal to the rate,
The command value achievement probability density distribution is estimated with respect to the maximum deceleration based on the vehicle stop characteristics of the vehicle, which is measured in advance by executing the stop operation by the deceleration specified for the vehicle on the vehicle. Was done,
The error distribution includes the sensor configuration and the detection distance from the distance detection characteristics of the distance detection unit, which are measured in advance using the distance detection unit in a situation where the true value of the distance to the target is known. The vehicle control device according to claim 2 or 3, wherein the detection time is estimated. The collision rate map is a grid point on the collision rate map _{(G xc,} G _vc) collision rate in _{p c (G xc, G vc} ) is the formula (I)

Obtained by, in formula (I),
p (d ^ ₁ ) is the detection rate at which the obstacle is detected by the distance detection unit at the grid points (G _xx , G _vc).
p (d ^ ₂ ) is the probability that the obstacle is detected at the position of the _{distance d ^ 2} by the distance detection unit at the grid points (G _xx , G _vc).
p _c (g _xn , g _cn ) transitions when the vehicle behavior control unit commands the vehicle to decelerate a so that the vehicle stops from the grid point (G _xc , G _{vc) to the target stop position.} Collision rate that will occur Collision rate at points (g _xn , g _cn ) on the map,
p (ε _α ) represents the probability that the _{uncertainty ε α,} which represents the difference from the deceleration actually executed by the vehicle with respect to the command of deceleration a, is taken.
_{The points (g xn} , g _cn ) on the collision rate map are points obtained by adding the uncertainty ε _{α to the deceleration a.}
The collision rate on the grid point where the velocity on the collision rate map is 0 is given based on the result of measuring the characteristics of the distance detection unit in advance, and the position on the collision rate map is the detection distance. The collision rate on the grid points other than velocity = 0 is given a predetermined value,
The value of p _c (g _xn , g _cn ) is obtained by an approximate calculation based on the collision rate at the grid points in the vicinity of the point (g _xn , g _cn).
The grid of the collision rate map is calculated by repeating the total calculation of the formula (I) in the direction in which the speed increases from the grid point corresponding to the speed = 0 and the position = the detection distance and / or in the direction in which the position approaches the vehicle from the detection distance. The vehicle control device according to claim 4, wherein a collision rate at a point (G _xc , G _{vc) is obtained.} The vehicle control device according to claim 5, wherein the probability p (ε _α ) has a normal distribution determined based on the characteristics of the vehicle measured in advance. The action plan generation unit is a distance detection unit in a situation where the result of a process of measuring a stop position when a certain deceleration is commanded to the vehicle and the true value of the distance to a target are known. It has a collision rate map storage unit that stores a plurality of collision rate maps obtained from the result of the process of measuring the characteristics of target detection by the above and the predetermined target collision rate.
The collision rate map setting unit
With the maximum deceleration
With the detection information
On the basis of,
Any one of claims 1 to 6, wherein one of the plurality of collision rate maps stored in the collision rate map storage unit is selected and determined as the collision rate map to be used. The vehicle control device according to the section. The collision rate map setting unit
The result of the process of measuring the stop position when a certain deceleration is ordered to the vehicle and the characteristics of the target detection by the distance detection unit in the situation where the true value of the distance to the target is known. The result of the measuring process and the parameters indicating the characteristics of the vehicle and the distance detector, which are derived from the measurement process,
With the predetermined target collision rate,
With the maximum deceleration
With the detection information
The vehicle control device according to any one of claims 1 to 6, wherein the collision rate map is determined by calculating in real time during traveling based on the above. The action plan generation unit has a collision rate map storage unit that stores the collision rate map calculated in real time by the collision rate map setting unit during traveling.
The collision rate map setting unit
With the predetermined target collision rate,
With the maximum deceleration
With the detection information
When the collision rate map corresponding to is stored in the collision rate map storage unit, it is determined as a collision rate map using the collision rate map stored in the collision rate map storage unit. The vehicle control device according to claim 8. The distance detection unit includes a plurality of sensors for detecting obstacles.
The detection information includes a sensor configuration indicating a combination of sensors that have detected the obstacle among the plurality of sensors, and a detection distance detected by the sensor configuration.
When the obstacle is detected only by a part of the plurality of sensors and the detection distance is equal to or more than a predetermined distance threshold, the action plan generation unit does not have the obstacle. Generate an action plan to carry out the cruising operation as the current action plan,
When the obstacle is detected only by a part of the plurality of sensors and the detection distance is smaller than the predetermined distance threshold, the action plan generation unit has the obstacle. The vehicle control device according to claim 1 or 2, wherein an action plan for performing preliminary braking as deemed to be assumed is generated as the current action plan. A method of controlling a vehicle by a vehicle behavior control unit that controls at least the speed of the vehicle based on an action plan for automatic driving of the vehicle.
The step of setting the maximum deceleration in the automatic driving of the vehicle and
A step of detecting the distance between the vehicle and an obstacle and obtaining detection information which is information related to the detection of the obstacle.
When the obstacle is detected, a step of determining a collision rate map, which is a two-dimensional map showing the distribution of the collision rate indicating the possibility of collision between the vehicle and the obstacle in a two-dimensional space of position and speed, and a step of determining the collision rate map.
A step of controlling at least the speed of the vehicle based on the determined collision rate map, a predetermined target collision rate, and the current position and speed of the vehicle.
Is a vehicle control method including
The detection information includes a sensor configuration indicating a combination of sensors that have detected an obstacle among a plurality of sensors, a detection distance detected by the sensor configuration, and detection that is continuously detected by the sensor configuration. Including time
The collision rate map is created on the premise of a target stop position set based on a combination of the predetermined target collision rate, the maximum deceleration, and the detection information.
The collision rate map defines a plurality of grid points in the two-dimensional space.
A collision rate is set for each grid point,
The collision rate at each grid point is determined by the vehicle behavior control unit with the maximum deceleration as the upper limit so that the vehicle decelerates from the position and speed corresponding to the grid point and stops at the target stop position. A vehicle control method, characterized in that the probability that the vehicle collides with the obstacle when a command is given to the vehicle is expressed.

Images