JP2018165692A - Driving support method and driving support device using the same, automatic driving control device, vehicle, program, and presentation system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a technology to derive a driving behavior that reflects a driver's intention.SOLUTION: A creation part 90 creates presentation information in which a plurality of types of driving behaviors are indicated. A presentation information output part outputs presentation information to a notification device. An operation signal input part receives an input of an operation signal indicating one driving behavior selected by a driver from the presentation information. A selection part 94, when the operation signal is not input, selects one driving behavior from the plurality of driving behaviors. A learning part 74, when the operation signal is input, executes weighting on one driving behavior indicated in the operation signal, and when the operation signal is not input, executes weighting on the one driving behavior selected by the selection part 94. The learning part 74 makes the value of the weighting when the operation signal is input larger than the value of the weighting when the operation signal is not input.SELECTED DRAWING: Figure 3

Description

  The present invention relates to a vehicle, a driving support method provided in the vehicle, a driving support device using the same, an automatic driving control device, a program, and a presentation system.

  An autonomous driving vehicle travels by detecting a situation around the vehicle and automatically executing a driving action that has been intended and performed by a passenger. Such an automatic driving vehicle is equipped with a driving support device for the occupant to change the driving behavior so that the driving behavior intended by the occupant and the driving behavior of the automatic driving vehicle do not deviate. The driving support device presents an executable driving action and causes the occupant to select the driving action (see, for example, Patent Document 1).

International Publication No. 16/170763

  If the presented driving behavior does not include the driving behavior desired by the occupant, the occupant cannot select the desired driving behavior. Therefore, it is desirable to present driving behavior that reflects the passenger's intention.

  This invention is made | formed in view of such a condition, The objective is to provide the technique which derives | leads-out the driving action which reflects a passenger | crew's intent.

  In order to solve the above-described problem, a driving support device according to an aspect of the present invention includes a generation unit that generates presentation information indicating a plurality of types of driving behavior that is an estimation result using a driving behavior model, and a generation unit A presentation information output unit for outputting the generated presentation information to the notification device, an operation signal input unit for inputting an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device, and an operation When the operation signal is not input to the signal input unit, when the operation signal is input to the selection unit that selects one driving action among a plurality of types of driving behavior and the operation signal input unit, the operation signal indicates If the driving behavior model is updated by performing learning while weighting one driving behavior, and no operation signal is input to the operation signal input unit, one operation selected by the selection unit is performed. And a learning unit for updating the driving behavior model by performing the learning while performing weighted behavior. The learning unit increases the weighting value when the operation signal is input to the operation signal input unit, than the weighting value when the operation signal is not input to the operation signal input unit.

  Another aspect of the present invention is an automatic operation control device. This device includes a generation unit that generates presentation information indicating a plurality of types of driving behavior that are estimation results using a driving behavior model, and a presentation information output unit that outputs the presentation information generated in the generation unit to a notification device. An operation signal input unit to which an operation signal indicating one driving action selected by the occupant is input with respect to the presentation information notified from the notification device, and a plurality of types when no operation signal is input to the operation signal input unit When the operation signal is input to the selection unit that selects one of the driving behaviors and the operation signal input unit, the vehicle is automatically driven based on the one driving behavior indicated in the operation signal. When an operation signal is not input to the operation signal input unit, an automatic operation control unit that controls automatic driving of the vehicle based on one driving action selected by the selection unit, and an operation signal to the operation signal input unit Is entered If the driving behavior model is updated by performing learning while weighting one driving behavior indicated in the operation signal, and the operation signal is not input to the operation signal input unit, the selection unit selects A learning unit that updates the driving behavior model by performing learning while weighting one driving behavior, and the learning unit is a weighting value when an operation signal is not input to the operation signal input unit. Rather, the weighting value when the operation signal is input to the operation signal input unit is increased.

  Yet another embodiment of the present invention is a vehicle. The vehicle includes a driving support device, and the driving support device generates a presentation information indicating a plurality of types of driving behaviors, which are estimation results using a driving behavior model, and a generation unit A presentation information output unit for outputting the generated presentation information to the notification device, an operation signal input unit for inputting an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device, and an operation When the operation signal is not input to the signal input unit, when the operation signal is input to the selection unit that selects one driving action among a plurality of types of driving behavior and the operation signal input unit, the operation signal indicates The driving behavior model is updated by executing learning while weighting one driving behavior, and when the operation signal is not input to the operation signal input unit, the one selected by the selection unit And a learning unit for updating the driving behavior model by performing the learning while weighting is performed on the rolling behavior. The learning unit increases the weighting value when the operation signal is input to the operation signal input unit, than the weighting value when the operation signal is not input to the operation signal input unit.

  Yet another embodiment of the present invention is a driving support method. In this method, a step of generating presentation information indicating a plurality of types of driving behaviors that are estimation results using a driving behavior model, a step of outputting the generated presentation information to a notification device, and a notification from the notification device A step of inputting an operation signal indicating one driving action selected by the occupant with respect to the presentation information; and a step of selecting one driving action among a plurality of types of driving actions when the operation signal is not input; When the operation signal is input, the step of updating the driving behavior model by executing learning while weighting one driving behavior indicated in the operation signal, and the operation signal is not input, Updating the driving behavior model by performing learning while weighting one driving behavior selected in the selecting step. . The weighting value when the operation signal is input is set larger than the weighting value when the operation signal is not input.

  Yet another embodiment of the present invention is a presentation system. This presentation system includes a generation unit that generates presentation information indicating a plurality of types of driving behaviors that are estimation results using a driving behavior model, and a presentation information output unit that outputs the presentation information generated by the generation unit. A driving support device; and a notification device that notifies the presentation information output from the driving support device. The driving support device includes an operation signal input unit that receives an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device, and no operation signal is input to the operation signal input unit. In this case, when an operation signal is input to the selection unit that selects one of the plurality of types of driving behavior and the operation signal input unit, weighting is performed on one driving behavior indicated in the operation signal. The driving behavior model is updated by executing learning while the operation signal is not input to the operation signal input unit, and driving is performed by performing learning while weighting one driving behavior selected by the selection unit. And a learning unit that updates the behavior model. The learning unit increases the weighting value when the operation signal is input to the operation signal input unit, than the weighting value when the operation signal is not input to the operation signal input unit.

  Arbitrary combinations of the above constituent elements, a part of the constituent elements, for example, a learning unit, are processed sequentially or collectively for a predetermined time such as one day by a computer via a communication network, and the expression of the present invention is an apparatus. A system, a method, a program, a recording medium on which the program is recorded, a vehicle in which the apparatus is mounted, and the like are also effective as an aspect of the present invention.

  According to the present invention, it is possible to derive a driving action that reflects the intention of the occupant.

It is a figure which shows the structure of the vehicle which concerns on Embodiment 1 thru | or 3. It is a figure which shows typically the interior of the vehicle which concerns on Embodiment 1 thru | or 3. 3 is a diagram illustrating a configuration of a control unit according to Embodiments 1 to 3. FIG. 6 is a diagram showing a histogram generated in the histogram generation unit according to Embodiments 1 to 3. FIG. FIGS. 5A to 5C are diagrams showing an outline of processing of the display control unit of FIG. 4 is a flowchart illustrating a processing procedure by a control unit according to the first embodiment. FIGS. 7A to 7E are diagrams illustrating an outline of processing of the display control unit according to the second embodiment. 10 is a flowchart illustrating a generation procedure by a control unit according to the second embodiment. 10 is a flowchart illustrating a generation procedure by a control unit according to Embodiment 3.

(Embodiment 1)
Before describing the present invention in detail, an outline will be described. The present embodiment relates to an automatic driving of an automobile. In particular, the present embodiment is a device that controls an HMI (Human Machine Interface) for exchanging information related to driving behavior of the vehicle with a vehicle occupant (for example, a driver) (hereinafter also referred to as “driving support device”). .) Various terms in the present embodiment are defined as follows. “Driving behavior” includes the state of operation such as steering and braking during driving or stopping of the vehicle, or control content related to automatic driving control, for example, constant speed driving, acceleration, deceleration, pause, stop, Lane change, course change, left / right turn, parking, etc. In addition, driving behavior includes cruise (maintaining lane keeping, vehicle speed), lane keeping, following vehicle follow-up, stop-and-go during follow-up, overtaking, response to merging vehicles, transfer including highway entry and exit ( Interchange), confluence, response to construction zone, emergency vehicle response, interrupt vehicle response, right / left turn lane response, pedestrian / bicycle interaction, obstacle avoidance other than vehicles, signs response , Right / left turn / U-turn restrictions, lane restrictions, one-way traffic, traffic signs, intersections / roundabouts, etc.

  As the “driving behavior estimation engine”, any of DL (Deep Learning), ML (Machine Learning), a filter, or a combination thereof is used. Deep Learning is, for example, CNN (Convolutional Neural Network) or RNN (Recurrent Neural Network). Also, the Machine Learning is, for example, SVM (Support Vector Machine). Furthermore, the filter is, for example, collaborative filtering.

  The “driving behavior model” is uniquely determined according to the driving behavior estimation engine. The driving behavior model in the case of DL is a learned neural network (Neural Network), the driving behavior model in the case of SVM is a learned prediction model, and the driving behavior model in the case of collaborative filtering includes driving environment data and driving It is data that links behavior data. The driving behavior model in the case of a rule is data in which an input and an output are linked.

  Based on such a definition, the driving support apparatus estimates a plurality of driving behaviors using a driving behavior model generated by machine learning or the like. Further, the driving support device selects the driving action having the highest reliability and causes the driving operation to be performed according to the selected driving action. The reliability indicates the certainty of the estimated driving behavior, corresponds to the cumulative value of the estimation result in the case of DL, corresponds to the confidence value in the case of SVM, and corresponds to the case of collaborative filtering. Corresponds to the degree of correlation. In the case of rules, this corresponds to the reliability of the rules. Therefore, the driving action with the highest reliability is a driving action with high safety. However, the driving behavior may not be a driving behavior that reflects the occupant's intention, and it is desirable to derive a driving behavior that reflects the occupant's intention.

  Therefore, in the present embodiment, the driving behavior model is updated by executing reinforcement learning based on the selected driving behavior. At this time, the reward for the case where the driving action is selected by the occupant is set larger than the reward for the case where the driving action is selected by the driving support device. As a result, the reliability of the driving action selected by the occupant is likely to increase thereafter, and the intention of the occupant is easily reflected. Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Each embodiment described below is an example, and the present invention is not limited to these embodiments.

  FIG. 1 shows a configuration of vehicle 100 according to Embodiment 1, and particularly shows a configuration related to an autonomous driving vehicle. The vehicle 100 can travel in the automatic driving mode, and includes a notification device 2, an input device 4, a wireless device 8, a driving operation unit 10, a detection unit 20, an automatic driving control device 30, and a driving support device 40. The devices shown in FIG. 1 may be connected by wired communication such as a dedicated line or a CAN (Controller Area Network). Moreover, you may connect by wired communication or wireless communications, such as USB (Universal Serial Bus), Ethernet (trademark), Wi-Fi (trademark), Bluetooth (trademark).

  The notification device 2 notifies the occupant of information related to traveling of the vehicle 100. The notification device 2 is, for example, a car navigation system, a head-up display, or a center display installed in the vehicle. The notification device 2 may be a display unit that displays information such as a light emitter such as an LED (Light Emitting Diode) installed around the steering wheel, pillar, dashboard, and meter panel. In addition, the notification device 2 may be a speaker that converts information into sound and notifies the occupant, or a vibration body provided at a position that can be sensed by the occupant (for example, the occupant's seat, steering wheel). There may be. Further, the notification device 2 may be a combination thereof.

  The input device 4 is a user interface device that receives an operation input by an occupant. For example, the input device 4 is a touch panel, a lever, a button, a switch, a controller such as a joystick or a volume, a sensor such as a camera that recognizes a gesture without contact, a sensor such as a microphone that recognizes voice, or a combination thereof. Receives information on the automatic driving of the vehicle entered by. An operation signal for switching between automatic operation and manual operation may be received. The input device 4 outputs the received information to the driving support device 40 as an operation signal.

  FIG. 2 schematically shows the interior of the vehicle 100. The notification device 2 may be a head-up display (HUD, Head-Up Display) 2a or a center display 2b. The input device 4 may be the first operation unit 4a provided in the steering 11 or the second operation unit 4b provided between the driver seat and the passenger seat, and is a camera that recognizes a gesture. The third operation unit 4c, which is a sensor such as In addition, the alerting | reporting apparatus 2 and the input device 4 may be integrated, for example, may be mounted as a touch panel display. The vehicle 100 may further be provided with a speaker 6 that presents information related to automatic driving to the occupant by voice. In this case, the driving support device 40 may cause the notification device 2 to display an image indicating information related to automatic driving, and output a sound indicating information related to automatic driving from the speaker 6 together with or instead of the information. Returning to FIG.

  The wireless device 8 corresponds to a mobile phone communication system, WMAN (Wireless Metropolitan Area Network), and the like, and performs wireless communication. Specifically, the wireless device 8 communicates with the server 300 via the network 302. Server 300 is a device external to vehicle 100 and includes a driving behavior learning unit 310. The driving behavior learning unit 310 will be described later. The server 300 and the driving support device 40 are included in the driving support system 500.

  The driving operation unit 10 includes a steering 11, a brake pedal 12, an accelerator pedal 13, and a winker switch 14. The steering 11, the brake pedal 12, the accelerator pedal 13, and the winker switch 14 can be electronically controlled by a steering ECU, a brake ECU, an engine ECU, a motor ECU, and a winker controller. In the automatic operation mode, the steering ECU, the brake ECU, the engine ECU, and the motor ECU drive the actuator in accordance with a control signal supplied from the automatic operation control device 30. The blinker controller turns on or off the blinker lamp according to a control signal supplied from the automatic operation control device 30.

  The detection unit 20 detects a surrounding situation and a running state of the vehicle 100. The detection unit 20 includes, for example, the speed of the vehicle 100, the relative speed of the preceding vehicle with respect to the vehicle 100, the distance between the vehicle 100 and the preceding vehicle, the relative speed of the vehicle in the side lane with respect to the vehicle 100, and the vehicle in the vehicle 100 and the side lane. And the position information of the vehicle 100 are detected. The detection unit 20 outputs various detected information (hereinafter referred to as “detection information”) to the automatic operation control device 30. Further, the detection unit 20 may output the detection information to the driving support device 40 via the automatic driving control device 30 or may directly output the detection information to the driving support device 40. The detection unit 20 includes a position information acquisition unit 21, a sensor 22, a speed information acquisition unit 23, and a map information acquisition unit 24.

  The position information acquisition unit 21 acquires the current position of the vehicle 100 from a GNSS (Global Navigation Satellite System (s)) receiver. The sensor 22 is a generic name for various sensors for detecting the situation outside the vehicle and the state of the vehicle 100. For example, a camera, a millimeter wave radar, a LIDAR (Light Detection and Ranging), a sonar, a temperature sensor, an atmospheric pressure sensor, a humidity sensor, and an illuminance sensor are mounted as sensors for detecting a situation outside the vehicle. The situation outside the vehicle includes a road condition in which the host vehicle includes lane information, an environment including weather, a situation around the host vehicle, and other vehicles in the vicinity (such as other vehicles traveling in the adjacent lane). Any information outside the vehicle that can be detected by the sensor 22 may be used. Further, as the sensor 22 for detecting the state of the vehicle 100, for example, an acceleration sensor, a gyro sensor, a geomagnetic sensor, an inclination sensor, and the like are mounted.

  The speed information acquisition unit 23 acquires the current speed of the vehicle 100 from the vehicle speed sensor. The map information acquisition unit 24 acquires map information around the current position of the vehicle 100 from the map database. The map database may be recorded on a recording medium in the vehicle 100, or may be downloaded from a map server via a network when used. The map information includes information on roads and intersections.

  The automatic driving control device 30 is an automatic driving controller that implements an automatic driving control function, and determines the behavior of the vehicle 100 in automatic driving. The automatic operation control device 30 includes a control unit 31, a storage unit 32, and an I / O (Input / Output, input / output) unit 33. The configuration of the control unit 31 can be realized by cooperation of hardware resources and software resources, or only by hardware resources. Processors, ROM (Read Only Memory), RAM (Random Access Memory), and other LSIs can be used as hardware resources, and programs such as operating systems, applications, and firmware can be used as software resources. The storage unit 32 includes a nonvolatile recording medium such as a flash memory. The I / O unit 33 executes communication control according to various communication formats. For example, the I / O unit 33 outputs information related to automatic driving to the driving support device 40 and inputs a control command from the driving support device 40. Further, the I / O unit 33 inputs detection information from the detection unit 20.

  The control unit 31 applies the control command input from the driving support device 40 and various information collected from the detection unit 20 or various ECUs to the automatic driving algorithm, and automatically controls the accelerator throttle opening, steering angle, etc. of the vehicle 100. A control value for controlling the object is calculated. The control unit 31 transmits the calculated control value to each control target ECU or controller. In this embodiment, it is transmitted to the steering ECU, the brake ECU, the engine ECU, and the winker controller. In the case of an electric vehicle or a hybrid car, the control value is transmitted to the motor ECU instead of or in addition to the engine ECU.

  The driving support device 40 is an HMI controller that executes an interface function between the vehicle 100 and an occupant, and includes a control unit 41, a storage unit 42, and an I / O unit 43. The control unit 41 executes various data processing such as HMI control. The control unit 41 can be realized by cooperation of hardware resources and software resources, or only by hardware resources. Processors, ROM, RAM, and other LSIs can be used as hardware resources, and programs such as an operating system, application, and firmware can be used as software resources.

  The storage unit 42 is a storage area that stores data that is referred to or updated by the control unit 41. For example, it is realized by a non-volatile recording medium such as a flash memory. The I / O unit 43 executes various communication controls according to various communication formats. The I / O unit 43 includes an operation signal input unit 50, an image / audio output unit 51, a detection information input unit 52, a command IF (Interface) 53, and a communication IF 56.

  The operation signal input unit 50 receives an operation signal from the input device 4 by an operation of an occupant or a user outside the vehicle made to the input device 4 and outputs the operation signal to the control unit 41. The image / sound output unit 51 outputs the image data or the voice message generated by the control unit 41 to the notification device 2 for display. The detection information input unit 52 is a result of the detection process by the detection unit 20, receives detection information indicating the current surrounding state and running state of the vehicle 100 from the detection unit 20, and outputs the detection information to the control unit 41.

  The command IF 53 executes interface processing with the automatic driving control device 30 and includes an action information input unit 54 and a command output unit 55. The behavior information input unit 54 receives information regarding the automatic driving of the vehicle 100 transmitted from the automatic driving control device 30 and outputs the information to the control unit 41. The command output unit 55 receives a control command for instructing the automatic driving control device 30 from the automatic driving control device 30 and transmits the control command to the automatic driving control device 30.

  The communication IF 56 executes interface processing with the wireless device 8. The communication IF 56 transmits the data output from the control unit 41 to the wireless device 8 and causes the wireless device 8 to transmit to the device outside the vehicle. Further, the communication IF 56 receives data from a device outside the vehicle transferred by the wireless device 8 and outputs the data to the control unit 41.

  Here, the automatic driving control device 30 and the driving support device 40 are configured as separate devices. As a modified example, as shown by a broken line in FIG. 1, the automatic driving control device 30 and the driving support device 40 may be integrated into one controller. In other words, one automatic driving control device may be configured to have the functions of both the automatic driving control device 30 and the driving support device 40 of FIG. Furthermore, you may be comprised as a presentation system with which the alerting | reporting apparatus 2 and the driving assistance device 40 were combined.

  FIG. 3 shows the configuration of the control unit 41. The control unit 41 includes a driving action estimation unit 70, a display control unit 72, and a learning unit 74. The driving behavior estimation unit 70 includes a driving behavior model 80, an estimation unit 82, and a histogram generation unit 84, the display control unit 72 includes a generation unit 90 and a processing unit 92, and the processing unit 92 includes a selection unit 94.

  The driving behavior estimation unit 70 uses a neural network (NN) that has been constructed in advance in order to determine driving behavior that can be realized in the current situation among a plurality of driving behavior candidates that the vehicle 100 can execute. To do. Here, there may be a plurality of driving behaviors that can be realized, and determining the driving behavior can be said to estimate the driving behavior.

  Since the processing in the driving behavior estimation unit 70 is also related to the driving behavior learning unit 310 in the server 300 in FIG. 1, here, the processing in the driving behavior learning unit 310 will be described first. The driving behavior learning unit 310 inputs at least one of driving histories and traveling histories of a plurality of drivers as a parameter to the neural network. The driving behavior learning unit 310 optimizes the weight of the neural network so that the output from the neural network matches the supervised data corresponding to the input parameter. The driving behavior learning unit 310 generates the driving behavior model 80 by repeatedly executing such processing. That is, the driving behavior model 80 is a neural network with optimized weights. The server 300 outputs the driving behavior model 80 generated by the driving behavior learning unit 310 to the driving support device 40 via the network 302 and the wireless device 8. Note that the driving behavior learning unit 310 may update the driving behavior model 80 based on a new parameter. At that time, the updated driving behavior model 80 may be output to the driving support device 40 in real time, or may be output to the driving support device 40 with a delay.

  The driving behavior model 80 generated by the driving behavior learning unit 310 and input to the driving behavior estimation unit 70 is a neural network constructed from at least one of a driving history and a driving history of a plurality of drivers. The driving behavior model 80 is a neural network in which a neural network constructed from the driving histories and traveling histories of a plurality of drivers is reconstructed by transfer learning using the traveling histories and traveling histories of specific drivers. May be. Since a known technique may be used for the construction of the neural network, the description is omitted here. 3 includes one driving behavior model 80. However, a plurality of driving behavior models 80 are provided for each driver, passenger, traveling scene, weather, and country. The status may be automatically determined or switched by manual change.

  The estimation unit 82 estimates driving behavior using the driving behavior model 80. Here, the driving history indicates a plurality of feature amounts (hereinafter referred to as “feature amount set”) corresponding to each of a plurality of driving actions performed by the vehicle 100 in the past. The plurality of feature amounts corresponding to the driving action are, for example, quantities indicating the driving state of the vehicle 100 at a time point a predetermined time before the driving action is performed by the vehicle 100. The feature amount is, for example, the number of passengers, the speed of the vehicle 100 and its time series, the steering amount and its time series of the steering wheel, the degree of brake and its time series, the degree of accelerator and its time series, and the like. The driving history may be referred to as a driving characteristic model. Therefore, the feature amount is, for example, a feature amount related to speed, a feature amount related to steering, a feature amount related to operation timing, a feature amount related to outside vehicle sensing, or a feature amount related to in-vehicle sensing. These feature amounts are detected by the detection unit 20 in FIG. 1 and input to the estimation unit 82 via the I / O unit 43. These feature amounts may be added to the driving history and the driving history of a plurality of drivers, and may be used for newly reconstructing a neural network. Furthermore, these feature amounts may be added to the travel history and travel history of a specific driver, and may be used for newly reconstructing a neural network.

  The travel history indicates a plurality of environmental parameters (hereinafter referred to as “environment parameter set”) corresponding to each of a plurality of driving actions performed by the vehicle 100 in the past. The plurality of environmental parameters corresponding to the driving behavior include, for example, the environment (surrounding conditions) of the vehicle 100 in a predetermined time range before a predetermined time from the time when the driving behavior is performed by the vehicle 100 or in a predetermined time range before that time. It is a parameter to show. The environmental parameters are, for example, the speed of the host vehicle, the relative speed of the preceding vehicle with respect to the host vehicle, the size at which the sensor detects the preceding vehicle, and the inter-vehicle distance between the preceding vehicle and the host vehicle. Further, these environmental parameters are detected by the detection unit 20 of FIG. 1 and input to the estimation unit 82 via the I / O unit 43. Further, these environmental parameters may be added to the traveling history and traveling history of a plurality of drivers, and may be used for newly reconstructing a neural network. Furthermore, these environmental parameters may be added to the travel history and travel history of a specific driver and used for reconstructing a neural network.

  The estimation unit 82 acquires a feature set or / and environmental parameters included in the driving history or the traveling history. The estimation unit 82 inputs a feature set or / and environmental parameters to the neural network of the driving behavior model 80, and outputs the output from the neural network to the histogram generation unit 84 as an estimation result.

  The histogram generation unit 84 acquires a plurality of types of driving behaviors and estimation results corresponding to the driving behaviors from the estimation unit 82, and generates a histogram indicating the cumulative value of the estimation results for the driving behaviors. Therefore, the histogram includes a plurality of types of driving behaviors and cumulative values corresponding to the driving behaviors. Here, the cumulative value is a value obtained by accumulating the number of times the estimation result for the driving action is derived.

  FIG. 4 shows a histogram generated by the histogram generator 84. The histogram includes driving actions A to E which are five kinds of driving actions as an example. In addition, a cumulative value for each of the driving actions A to E is included. Here, it is assumed that the driving action C, the driving action E, the driving action B, the driving action D, and the driving action A are in descending order of the cumulative value. Returning to FIG. The histogram generation unit 84 outputs the generated histogram to the generation unit 90.

  The generation unit 90 inputs a histogram from the histogram generation unit 84, that is, a plurality of types of driving behaviors, and a cumulative value corresponding to each driving behavior. The generation unit 90 selects a predetermined number of driving actions in descending order of the cumulative value of each of the plurality of types of driving actions. For example, the generation unit 90 selects the driving action “3” from the driving actions “5”. Note that the number of selected driving actions is not limited to “3”. It can be said that the selected driving behavior is part of the plurality of types of driving behavior from the histogram generation unit 84. In addition, when the preceding vehicle decelerates, it may be limited to safe driving behavior estimated based on a rule base in line with traffic safety, such as excluding driving behavior that creates a danger of accelerating and closing the distance between vehicles . The generation unit 90 generates presentation information indicating the selected driving action. In the case of FIG. 4, the presentation information includes driving behavior C, driving behavior E, and driving behavior B, and in the presentation information, these driving behaviors are arranged in descending order of the cumulative value. The generation unit 90 outputs the generated presentation information to the processing unit 92.

  The processing unit 92 receives presentation information from the generation unit 90. The processing unit 92 outputs the presentation information to the head-up display 2a or the center display 2b in FIG. 2 via the image / sound output unit 51 in FIG. The head-up display 2a or the center display 2b displays an image of presentation information. The processing unit 92 may output the presentation information to the speaker 6 in FIG. 2 via the image / sound output unit 51 in FIG. At that time, the speaker 6 outputs a voice message of the presentation information.

  5A to 5C show an outline of processing of the display control unit 72. FIG. FIG. 5A shows an image of presentation information displayed on the center display 2b. The image of the presentation information is generated based on FIG. 4, and the driving action C, the driving action E, and the driving action B are arranged in order from the top to the bottom in the descending order of the cumulative value. Also, the larger the accumulated value, the larger the character size. That is, in the image displayed on the center display 2b, the driving action arranged forward in the presentation information is arranged at the top of the screen and the character size is increased. These are for the driving behavior with higher reliability, which is a cumulative value, to be more easily selected by the occupant.

  In addition, when the image of presentation information is displayed on the head up display 2a, the said image is the same as that of Fig.5 (a). FIG. 5B is an image of presentation information displayed on the center display 2b, and an image of presentation information when driving behavior C, driving behavior E, and driving behavior B are associated with actual driving behavior. Show. Here, it is assumed that the driving action C corresponds to “straight ahead”, the driving action E corresponds to “turn right”, and the driving action B corresponds to “change lane to left”. In the following, for clarity of explanation, FIG. The description of FIG. 5C will be described later, and the description returns to FIG.

  An occupant, for example, a driver, inputs a selection result to the input device 4 when selecting one driving action from a plurality of types of driving actions shown in the image of the presentation information displayed on the center display 2b. For example, the occupant selects one driving action by operating the first operation unit 4a or the second operation unit 4b. When the center display 2b is a touch panel, the occupant selects one driving action by touching a display portion of one driving action from the image of the presentation information displayed on the center display 2b. Furthermore, when the image of the presentation information is displayed on the head-up display 2a, when the occupant executes a gesture for selecting one display portion of the driving action from the image, the third operation unit 4c One driving action corresponding to the is selected. An operation signal that is an operation signal from the input device 4 and that indicates one driving action selected by the occupant is input to the operation signal input unit 50. When an operation signal is input to the operation signal input unit 50 in this way, it is called a “manual selection state”.

  When the occupant does not select one driving action from a plurality of types of driving actions shown in the image of the presentation information displayed on the center display 2b, the occupant does not input the selection result to the input device 4. Therefore, the operation signal input unit 50 does not receive an operation signal from the input device 4 for a predetermined period after the presentation action is output. When the operation signal is not input to the operation signal input unit 50, the processing unit 92 instructs the selection unit 94 to execute selection of one driving action. When the selection unit 94 receives an instruction from the processing unit 92, the selection unit 94 selects one driving behavior among a plurality of types of driving behaviors included in the presentation information. Here, for example, the exercise action having the largest cumulative value is selected. This is equivalent to selecting the first driving action among a plurality of types of driving actions arranged in the presentation information. When no operation signal is input to the operation signal input unit 50 as described above, this is called an “automatic selection state”.

  In the automatic selection state, the processing unit 92 outputs a control command corresponding to the selected driving action to the automatic driving control device 30 via the command output unit 55. On the other hand, in the manual selection state, the processing unit 92 outputs a control command corresponding to the driving behavior indicated by the operation signal to the automatic driving control device 30 via the command output unit 55. The automatic driving control device 30 in FIG. 1 controls the automatic driving of the vehicle 100 based on the driving behavior corresponding to the control command.

  In the automatic selection state, the learning unit 74 gives a positive reward “α” to one driving action selected by the selection unit 94. On the other hand, in the manual selection state, the learning unit 74 gives a positive reward “β” to one driving action indicated in the operation signal. Here, the value of the positive reward “β” given in the case of the manual selection state is made larger than the value of the positive reward “α” given in the case of the automatic selection state. For example, the former is “+0.5” and the latter is “+0.7”. The learning unit 74 updates the driving behavior model 80 by executing reinforcement learning while giving a reward to one driving behavior. Since a known technique may be used for reinforcement learning, a description thereof is omitted here, but this corresponds to adjustment of the weight of the neural network in the driving behavior model 80 so that the reward is reflected.

  Here, in the case of the manual selection state, the learning unit 74 is not selected by one of the driving behaviors other than the one driving behavior indicated in the operation signal among the plurality of types of driving behaviors included in the presentation information, that is, by the occupant. A negative reward may be given to driving behavior. In that case, a negative reward may not be given to the driving action which was not included in presentation information, and may be given. Here, the negative reward is indicated as “−β”, but may be other values. The learning unit 74 performs reinforcement learning while also giving a negative reward. On the other hand, in the automatic selection state, the learning unit 74 does not give a negative reward to driving behaviors other than the one driving behavior selected by the selection unit 94.

  Here, the reward given in the learning part 74 is demonstrated still in detail, using FIG.5 (c). Here, it is assumed that an image of presentation information as shown in FIG. 5A is displayed as a premise. The pattern “1” in FIG. 5C corresponds to the automatic selection state. If the occupant does not select the driving action, the selection unit 94 selects the driving action C. As a result, the learning unit 74 gives a positive reward “+ α” to the driving action C. On the other hand, patterns “2” to “4” in FIG. 5C correspond to the manual selection state. When the occupant selects the driving action C in the pattern “2”, the learning unit 74 gives a positive reward “+ β” to the driving action C. In the pattern “3”, when the occupant selects the driving action E, the learning unit 74 gives a positive reward “+ β” to the driving action E and gives a negative reward “−β” to the driving action C. In the pattern “4”, when the occupant selects the driving action B, the learning unit 74 gives a positive reward “+ β” to the driving action B, and gives a negative reward “−β” to the driving actions C and E. To do.

  The operation of the driving support device 40 having the above configuration will be described. FIG. 6 is a flowchart illustrating a processing procedure performed by the control unit 41. The estimation unit 82 estimates a plurality of driving behaviors (S10). The generation unit 90 selects a predetermined driving action in descending order of the cumulative value (S12). The notification device 2 displays the presentation information (S14). When an operation signal is input to the operation signal input unit 50 (Y in S16), the learning unit 74 gives a positive reward “+ β” to the driving action indicated in the operation signal (S18), A negative reward “−β” is given to a driving action having a cumulative location larger than that of the driving action indicated by the operation signal in another driving action (S20). When the operation signal is not input to the operation signal input unit 50 (N in S16), the selection unit 94 selects the driving action having the maximum accumulated value (S22). The learning unit 74 gives a positive reward “+ α” to the selected driving action (S24).

  According to the present embodiment, since the reward value when the operation signal is input is larger than the reward value when the operation signal is not input, the reliability of the driving behavior when the occupant actively selects Can increase the sex. In addition, since the reliability of the driving action when the occupant actively selects, the driving action that reflects the occupant's intention can be derived. In addition, since a negative reward is given to the driving behavior other than the driving behavior selected by the occupant, the reliability of the driving behavior not selected by the occupant can be lowered. In addition, since the system gives a negative reward to the driving behavior that the system has estimated to be more reliable than the driving behavior selected by the occupant, the driving behavior selected by the occupant is estimated to be highly reliable. The difference between the reliability of the vehicle and the reliability of the driving behavior not selected by the occupant can be increased. In addition, since the difference between the reliability of the driving behavior selected by the occupant and the reliability of the driving behavior that the occupant did not select is estimated that the system is highly reliable, the will of the occupant can be further reflected.

  Moreover, since a negative reward is given to the driving action which was not selected among some driving actions contained in presentation information, it is not possible to give a reward to the driving action which was not included in presentation information. Moreover, since a reward is not given to the driving | running action which was not contained in presentation information, it can suppress that reliability changes irrespective of a passenger | crew's intent. Further, since the value of the positive reward when the operation signal is input is made larger than the value of the positive reward when the operation signal is not input, it is possible to execute a driving action that reflects the occupant's intention.

(Embodiment 2)
Next, a second embodiment will be described. As in the first embodiment, the second embodiment relates to a driving support device that displays an image of presentation information and performs reinforcement learning while rewarding a selected driving action. In the first embodiment, the reinforcement learning is executed by changing the value of the reward depending on whether the state is the automatic selection state or the manual selection state. On the other hand, in the second embodiment, the case where the vehicle is in the manual operation state will be described. As described above, the presentation information includes a predetermined number of driving actions among the estimated plural kinds of driving actions. The occupant can select a driving action included in the presentation information, but cannot select a driving action not included in the presentation information. Since the reliability of such driving behavior is not improved, the driving behavior is difficult to be estimated. In other words, since the occupant is excluded even if such driving behavior is not intentionally excluded, the intention of the occupant is not reflected. Embodiment 2 demonstrates the production | generation of the presentation information which becomes easy to reflect a passenger | crew's intent. The vehicle 100 according to the second embodiment is the same type as that shown in FIGS. 1 and 2, and the control unit 41 according to the second embodiment is the same type as that shown in FIG.

  The driving action estimation unit 70 in FIG. 3 outputs a histogram including a plurality of types of driving actions and cumulative values corresponding to the driving actions, as in the first embodiment. The generation unit 90 selects a predetermined number of driving behaviors in descending order of the cumulative value of each of a plurality of types of driving behaviors included in the histogram. The generation unit 90 generates presentation information indicating the selected driving action. This presentation information is the same as that in the first embodiment, but is referred to as “first presentation information” here. The first presentation information includes, for example, the driving action “3” from the larger cumulative value among the driving actions “5”.

  On the other hand, the generation unit 90 generates the second presentation information instead of the first presentation information at a predetermined frequency, for example, once every ten times. In the second presentation information, a driving action with a small cumulative value is included instead of a part of the predetermined number of driving actions to be included in the first presentation information. More specifically, when the predetermined number is “3”, the three driving actions to be included in the first presentation information are the first driving action having the cumulative value and the second driving action having the second cumulative value. The movement behavior and the cumulative value are the third driving behavior. The generation unit 90 is one of these driving actions, for example, the driving action having the fourth cumulative value, or the driving value having the fourth cumulative value, instead of the third driving action having the cumulative value. Second presentation information is generated so as to include the fifth driving action. For example, the driving behavior with the fourth cumulative value and the driving behavior with the fifth cumulative value may be included alternately at a predetermined ratio every time the second presentation information is generated. The generation unit 90 outputs the generated first presentation information or second presentation information to the processing unit 92.

  The processing unit 92 outputs the first presentation information or the second presentation information to the image / sound output unit 51 in FIG. 1, and the image / sound output unit 51 outputs the first presentation information to the head-up display 2a or the center display 2b in FIG. Presentation information or second presentation information is output. The head-up display 2a or the center display 2b displays an image of the first presentation information or the second presentation information.

  FIGS. 7A to 7E show an outline of processing of the display control unit 72 according to the second embodiment. FIG. 7A shows an image of the first presentation information displayed on the center display 2b. This is the same as FIG. FIG. 7B shows an image of the second presentation information displayed on the center display 2b. In order to facilitate comparison with the first presentation information, the image of the second presentation information is generated based on FIG. The driving action C and the driving action E are arranged side by side from top to bottom in order of increasing cumulative value. In addition, the driving action A having the smallest cumulative value is arranged at the bottom and has the smallest character size. 7 (c)-(e) will be described later, and the description returns to FIG.

  When the occupant does not select one driving action from a plurality of types of driving actions shown in the image of the first presentation information or the second presentation information displayed on the center display 2b, the occupant inputs a selection result to the input device 4. do not do. This is the aforementioned automatic selection state, and the operation signal from the input device 4 is not input to the operation signal input unit 50. In that case, the selection unit 94 selects one driving action among a plurality of types of driving actions included in the first presentation information or the second presentation information. More specifically, the selection unit 94 has a predetermined order among the predetermined number of driving actions included in the first presentation information or the second presentation information, for example, the one arranged at the top in the image. Select driving behavior. Such processing of the selection unit 94 is the same as that of the first embodiment. Considering such an operation of the selection unit 94, when generating the second presentation information, the generation unit 90 can be said to arrange driving actions having a small cumulative value in an order other than the order selected by the selection unit 94. . In FIG. 7B, the driving action having the smallest cumulative value is arranged at the bottom.

  On the other hand, when the occupant selects one driving action from a plurality of types of driving actions shown in the image of the first presentation information or the second presentation information displayed on the center display 2b, a selection result is input to the input device 4. Enter. This is the above-described manual selection state, and the operation signal input unit 50 receives an operation signal from the input device 4. The operation signal when the first presentation information is displayed is referred to as a “first operation signal”, and the operation signal when the second presentation information is displayed may be referred to as a “second operation signal”. Subsequently, the processing unit 92 outputs a control command to the automatic operation control device 30 via the command output unit 55, but since it is the same as in the first embodiment, the description thereof is omitted here.

  The learning unit 74 performs the same processing as in the first embodiment regardless of whether the first presentation information is output or the second presentation information is output in the automatic selection state. To do. Further, the learning unit 74 gives a reward to one driving action selected by the occupant when the first operation signal is input in the manual selection state. This reward is also given in the same manner as in the first embodiment, and a positive reward is given to one driving action indicated in the first operation signal. However, the learning unit 74 does not give a negative reward to a driving action that has not been selected among a plurality of types of driving actions included in the first presentation information in the manual selection state.

  When the second operation signal is input in the manual selection state, the learning unit 74 gives a reward to one driving action selected by the occupant. This reward is also given in the same manner as in the first embodiment, and a positive reward is given to one driving action indicated in the second operation signal. On the other hand, in the manual selection state, the learning unit 74 gives a negative reward to a driving action that is not selected among a plurality of types of driving actions included in the second presentation information. Here, the negative reward is indicated as “−β”, but may be other values.

  Here, the reward given in the learning part 74 is demonstrated still in detail, using FIG.7 (c)-(d). Here, only the manual selection state will be described. Assume that an image of the first presentation information as shown in FIG. 7A is displayed as a premise of FIG. In the pattern “1”, when the occupant selects the driving action C, the learning unit 74 gives a positive reward “+ β” to the driving action C. In the pattern “2”, when the occupant selects the driving action E, the learning unit 74 gives a positive reward “+ β” to the driving action E. In the pattern “3”, when the occupant selects the driving action B, the learning unit 74 gives the driving action B a positive reward “+ β”.

  Further, as an assumption of FIG. 7D, it is assumed that an image of the second presentation information as shown in FIG. 7B is displayed. In the pattern “4”, when the occupant selects the driving action C, the learning unit 74 gives a positive reward “+ β” to the driving action C. In the pattern “5”, when the occupant selects the driving action E, the learning unit 74 gives a positive reward “+ β” to the driving action E and gives a negative reward “−β” to the driving actions C and A. To do. In the pattern “6”, when the occupant selects the driving action A, the learning unit 74 gives a positive reward “+ β” to the driving action A, and gives a negative reward “−β” to the driving actions C and E. To do. Returning to FIG.

  When generating the second presentation information, the generation unit 90 so far arranges driving behaviors with small cumulative values in an order other than the order selected by the selection unit 94. Note that an example of the order of selection in the selection unit 94 is the highest in the image. However, when generating the second presentation information, the generation unit 90 may arrange driving actions having a small cumulative value in the order selected by the selection unit 94. FIG. 7E shows an image of the second presentation information displayed on the center display 2b in such a case. In order to facilitate comparison with the past, the image of the second presentation information is generated based on FIG. The driving action A having the smallest cumulative value is arranged at the top, and the size of the largest character is set. Also, below this, the driving action C and the driving action E are arranged in order from the top to the bottom in the order of the cumulative value.

  The operation of the driving support device 40 having the above configuration will be described. FIG. 8 is a flowchart showing a generation procedure by the control unit 41 according to the second embodiment. The generation unit 90 sets i = 1 (S100). If i is not 10 (N in S102), the generation unit 90 generates first presentation information (S104). The generation unit 90 increments i (S106). If i = 10 (Y in S102), the generation unit 90 generates second presentation information (S108). The generation unit 90 sets i = 1 (S110). If not completed (N in S112), the process returns to Step 102. If not finished (Y in S112), the process is finished.

  FIG. 9 is a flowchart illustrating a generation procedure by the control unit 41 according to the third embodiment. The first operation signal or the second operation signal is input to the operation signal input unit 50 (S150). When the first operation signal is input (Y in S152), the learning unit 74 gives a positive reward “+ β” to the selected driving action (S154). When the first presentation information is not input (N in S152), the learning unit 74 gives a positive reward “+ β” to the selected driving action (S156), and the selected driving action among the other driving actions is selected. A negative reward “−β” is given to the driving behavior estimated by the system with higher reliability than the behavior (S158).

  According to the present embodiment, since the second presentation information including the driving behavior with low reliability is output at a predetermined frequency, it is possible to increase the selection opportunities for the driving behavior with low reliability. Moreover, since the selection opportunity of driving behavior with low reliability increases, it can be distinguished whether or not the driving behavior was not selected by the occupant's intention. Further, since it is distinguished whether or not the driving behavior is not selected by the occupant's intention, the occupant's intention can be recognized. In addition, since the occupant's intention is recognized, driving behavior that reflects the occupant's intention can be derived. In addition, since the driving behavior with low reliability is arranged as the driving behavior selected by the selection unit, it is possible to recognize the positive will of the occupant depending on whether the occupant selects the driving behavior. In addition, since the driving behavior with low reliability is arranged other than the driving behavior selected in the selection unit, the driving behavior with high reliability can be selected by the selection unit. In addition, since the second presentation information including the driving behavior with low reliability is output at a predetermined frequency, the driving behavior reflecting the passenger's intention can be executed.

  As mentioned above, although embodiment concerning this invention has been explained in full detail with reference to drawings, the function of the apparatus mentioned above and each processing part may be realized by a computer program. A computer that realizes the above-described functions by a program includes an input device such as a keyboard, mouse, and touch pad, an output device such as a display and a speaker, a CPU (Central Processing Unit), a ROM, a RAM, a hard disk device, and an SSD (Solid State Drive). Storage device such as a DVD-ROM (Digital Versatile Disk Read Only Memory), a reading device that reads information from a recording medium such as a USB memory, a network card that communicates via a network, etc., and each part is connected by a bus .

  The reading device reads the program from the recording medium on which the program is recorded, and stores the program in the storage device. Or a network card communicates with the server apparatus connected to the network, and memorize | stores the program for implement | achieving the function of said each apparatus downloaded from the server apparatus in a memory | storage device. Further, the function of each device is realized by the CPU copying the program stored in the storage device to the RAM and sequentially reading out and executing the instructions included in the program from the RAM.

The outline of one embodiment of the present invention is as follows.
(Item 1-1)
A generation unit that generates presentation information indicating a plurality of types of driving behavior, which is an estimation result using the driving behavior model;
A presentation information output unit that outputs the presentation information generated in the generation unit to a notification device;
An operation signal input unit that receives an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device;
When an operation signal is not input to the operation signal input unit, a selection unit that selects one driving action among a plurality of types of driving actions;
When an operation signal is input to the operation signal input unit, the driving behavior model is updated by performing learning while weighting one driving behavior indicated in the operation signal, and the operation signal input unit A learning unit that updates a driving behavior model by performing learning while weighting one driving behavior selected in the selection unit when an operation signal is not input;
The learning unit increases a weighting value when an operation signal is input to the operation signal input unit, than a weighting value when an operation signal is not input to the operation signal input unit. Driving assistance device.

  According to this aspect, since the weighting value when the operation signal is input is made larger than the weighting value when the operation signal is not input, it is possible to derive driving behavior that reflects the occupant's intention.

(Item 1-2)
When the operation signal is input to the operation signal input unit, the learning unit performs learning while performing negative weighting on a driving action other than one driving action indicated in the operation signal. The driving support device according to item 1-1.
In this case, since negative weighting is performed on driving actions other than one driving action selected by the occupant, the reliability of driving actions not selected by the occupant can be reduced.

(Item 1-3)
The generation unit generates presentation information indicating some driving behaviors among a plurality of types of driving behaviors,
The learning unit, when an operation signal is input to the operation signal input unit, out of a part of driving behavior included in the presentation information generated in the generation unit, one driving behavior indicated in the operation signal The driving support device according to item 1-2, wherein negative weighting is performed on driving behavior other than the above.
In this case, since the negative weighting is performed on the driving behavior that has not been selected among some of the driving behaviors included in the presentation information, the reliability of the driving behavior not selected by the occupant can be reduced.

(Item 1-4)
A generation unit that generates presentation information indicating a plurality of types of driving behavior, which is an estimation result using the driving behavior model;
A presentation information output unit that outputs the presentation information generated in the generation unit to a notification device;
An operation signal input unit that receives an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device;
When an operation signal is not input to the operation signal input unit, a selection unit that selects one driving action among a plurality of types of driving actions;
When an operation signal is input to the operation signal input unit, automatic driving of the vehicle is controlled based on one driving action indicated in the operation signal, and no operation signal is input to the operation signal input unit An automatic driving control unit that controls automatic driving of the vehicle based on one driving action selected by the selection unit;
When an operation signal is input to the operation signal input unit, the driving behavior model is updated by performing learning while weighting one driving behavior indicated in the operation signal, and the operation signal input unit A learning unit that updates a driving behavior model by performing learning while weighting one driving behavior selected in the selection unit when an operation signal is not input;
The learning unit increases a weighting value when an operation signal is input to the operation signal input unit, than a weighting value when an operation signal is not input to the operation signal input unit. Automatic operation control device.

  According to this aspect, since the weighting value when the operation signal is input is made larger than the weighting value when the operation signal is not input, it is possible to execute the driving action that reflects the occupant's intention.

(Item 1-5)
A vehicle equipped with a driving support device,
The driving support device includes:
A generation unit that generates presentation information indicating a plurality of types of driving behavior, which is an estimation result using the driving behavior model;
A presentation information output unit that outputs the presentation information generated in the generation unit to a notification device;
An operation signal input unit that receives an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device;
When an operation signal is not input to the operation signal input unit, a selection unit that selects one driving action among a plurality of types of driving actions;
When an operation signal is input to the operation signal input unit, the driving behavior model is updated by performing learning while weighting one driving behavior indicated in the operation signal, and the operation signal input unit A learning unit that updates a driving behavior model by performing learning while weighting one driving behavior selected in the selection unit when an operation signal is not input;
The learning unit increases a weighting value when an operation signal is input to the operation signal input unit, than a weighting value when an operation signal is not input to the operation signal input unit. Vehicle.

  According to this aspect, since the weighting value when the operation signal is input is made larger than the weighting value when the operation signal is not input, it is possible to derive driving behavior that reflects the occupant's intention.

(Item 1-6)
Generating presentation information indicating a plurality of types of driving behavior that is an estimation result using the driving behavior model;
Outputting the generated presentation information to a notification device;
An operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device; and
When an operation signal is not input, a step of selecting one driving action among a plurality of types of driving actions;
When an operation signal is input, updating the driving behavior model by performing learning while weighting one driving behavior indicated in the operation signal; and
A step of updating the driving behavior model by performing learning while weighting the one driving behavior selected in the selecting step when the operation signal is not inputted,
A driving support method, wherein a weighting value when an operation signal is input is made larger than a weighting value when an operation signal is not input.

(Item 1-7)
Generating presentation information indicating a plurality of types of driving behavior that is an estimation result using the driving behavior model;
Outputting the generated presentation information to a notification device;
An operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device; and
When an operation signal is not input, a step of selecting one driving action among a plurality of types of driving actions;
When an operation signal is input, updating the driving behavior model by performing learning while weighting one driving behavior indicated in the operation signal; and
A step of updating the driving behavior model by performing learning while weighting the one driving behavior selected in the selecting step when the operation signal is not inputted,
A program for causing a computer to execute a weighting value when an operation signal is input larger than a weighting value when an operation signal is not input.

(Item 1-8)
A driving support apparatus comprising: a generation unit that generates presentation information indicating a plurality of types of driving behaviors that are estimation results using a driving behavior model; and a presentation information output unit that outputs the presentation information generated in the generation unit; ,
A notification device that notifies the presentation information output from the driving support device;
The driving support device includes:
An operation signal input unit that receives an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device;
When an operation signal is not input to the operation signal input unit, a selection unit that selects one driving action among a plurality of types of driving actions;
When an operation signal is input to the operation signal input unit, the driving behavior model is updated by performing learning while weighting one driving behavior indicated in the operation signal, and the operation signal input unit A learning unit that updates the driving behavior model by performing learning while weighting one driving behavior selected by the selection unit when the operation signal is not input;
The learning unit increases a weighting value when an operation signal is input to the operation signal input unit, than a weighting value when an operation signal is not input to the operation signal input unit. To present system.

(Item 2-1)
A predetermined number of driving actions are selected in descending order of the reliability of each of a plurality of types of driving actions, which are estimation results using the driving action model, and first presentation information indicating the selected predetermined number of driving actions is generated A generator to
A presentation information output unit that outputs the first presentation information generated in the generation unit to a notification device;
An operation signal input unit that receives a first operation signal indicating one driving action selected by the occupant with respect to the first presentation information notified from the notification device;
A learning unit that updates a driving behavior model by performing learning while performing positive weighting on one driving behavior indicated in the first operation signal input to the operation signal input unit;
The generating unit, instead of a part of the predetermined number of driving actions to be included in the first presentation information, substitutes the second presentation information including driving actions with low reliability in place of the first presentation information at a predetermined frequency. Generate and
The presentation information output unit outputs the second presentation information generated by the generation unit to the notification device,
The operation signal input unit receives a second operation signal indicating one driving action selected by the occupant with respect to the second presentation information notified from the notification device,
The learning unit performs positive weighting on one driving action indicated in the second operation signal input to the operation signal input unit, and is negative with respect to other driving actions included in the second presentation information. A driving support apparatus that updates a driving behavior model by performing learning while performing weighting.

  According to this aspect, since the second presentation information including the driving behavior with low reliability is output at a predetermined frequency, it is possible to derive the driving behavior reflecting the passenger's intention.

(Item 2-2)
When the first operation signal is not input to the operation signal input unit, one driving action included in the first presentation information is selected, and when the second operation signal is not input to the operation signal input unit, A selection unit for selecting one driving action included in the second presentation information;
The selection unit selects one driving action arranged in a predetermined order from among a predetermined number of driving actions included in the first presentation information or the second presentation information,
The driving support device according to Item 2-1, wherein the generation unit arranges driving actions with low reliability in the predetermined order when generating the second presentation information.
In this case, since the driving behavior with low reliability is arranged as the driving behavior selected by the selection unit, it is possible to recognize the positive will of the occupant.

(Item 2-3)
When the first operation signal is not input to the operation signal input unit, one driving action included in the first presentation information is selected, and when the second operation signal is not input to the operation signal input unit, A selection unit for selecting one driving action included in the second presentation information;
The selection unit selects one driving action arranged in a predetermined order from among a predetermined number of driving actions included in the first presentation information or the second presentation information,
The driving support device according to Item 2-1, wherein the generating unit arranges driving behaviors with low reliability in an order other than the predetermined order when generating the second presentation information.
In this case, since the driving behavior with low reliability is arranged in addition to the driving behavior selected by the selection unit, the driving behavior with high reliability can be selected by the selection unit.

(Item 2-4)
A predetermined number of driving actions are selected in descending order of the reliability of each of a plurality of types of driving actions, which are estimation results using the driving action model, and first presentation information indicating the selected predetermined number of driving actions is generated A generator to
A presentation information output unit that outputs the first presentation information generated in the generation unit to a notification device;
An operation signal input unit that receives a first operation signal indicating one driving action selected by the occupant with respect to the first presentation information notified from the notification device;
A learning unit that updates a driving behavior model by performing learning while performing positive weighting on one driving behavior indicated in the first operation signal input to the operation signal input unit;
With an automatic driving control unit that controls the automatic driving of the vehicle based on one driving action,
The generating unit, instead of a part of the predetermined number of driving actions to be included in the first presentation information, substitutes the second presentation information including driving actions with low reliability in place of the first presentation information at a predetermined frequency. Generate and
The presentation information output unit outputs the second presentation information generated by the generation unit to the notification device,
The operation signal input unit receives a second operation signal indicating one driving action selected by the occupant with respect to the second presentation information notified from the notification device,
The learning unit performs positive weighting on one driving action indicated in the second operation signal input to the operation signal input unit, and is negative with respect to other driving actions included in the second presentation information. An automatic driving control device that updates a driving behavior model by performing learning while executing weighting.

  According to this aspect, since the second presentation information including the driving action with low reliability is output at a predetermined frequency, the driving action reflecting the passenger's intention can be executed.

(Item 2-5)
A vehicle equipped with a driving support device,
The driving support device includes:
A predetermined number of driving actions are selected in descending order of the reliability of each of a plurality of types of driving actions, which are estimation results using the driving action model, and first presentation information indicating the selected predetermined number of driving actions is generated A generator to
A presentation information output unit that outputs the first presentation information generated in the generation unit to a notification device;
An operation signal input unit that receives a first operation signal indicating one driving action selected by the occupant with respect to the first presentation information notified from the notification device;
A learning unit that updates a driving behavior model by performing learning while performing positive weighting on one driving behavior indicated in the first operation signal input to the operation signal input unit;
The generating unit, instead of a part of the predetermined number of driving actions to be included in the first presentation information, substitutes the second presentation information including driving actions with low reliability in place of the first presentation information at a predetermined frequency. Generate and
The presentation information output unit outputs the second presentation information generated by the generation unit to the notification device,
The operation signal input unit receives a second operation signal indicating one driving action selected by the occupant with respect to the second presentation information notified from the notification device,
The learning unit performs positive weighting on one driving action indicated in the second operation signal input to the operation signal input unit, and is negative with respect to other driving actions included in the second presentation information. A vehicle that updates a driving behavior model by performing learning while performing weighting.

  According to this aspect, since the second presentation information including the driving behavior with low reliability is output at a predetermined frequency, it is possible to derive the driving behavior reflecting the passenger's intention.

(Item 2-6)
A predetermined number of driving actions are selected in descending order of the reliability of each of a plurality of types of driving actions, which are estimation results using the driving action model, and first presentation information indicating the selected predetermined number of driving actions is generated And steps to
Outputting the generated first presentation information to a notification device;
A step of inputting a first operation signal indicating one driving action selected by the occupant with respect to the first presentation information notified from the notification device;
Updating the driving behavior model by performing learning while performing positive weighting on one driving behavior indicated in the input first operation signal;
Generating a second presentation information including a driving action with low reliability instead of a part of the predetermined number of driving actions to be included in the first presentation information at a predetermined frequency instead of the first presentation information;
Outputting the generated second presentation information to the notification device;
A step of inputting a second operation signal indicating one driving action selected by the occupant with respect to the second presentation information notified from the notification device;
By performing positive weighting on one driving action indicated in the input second operation signal and performing learning while performing negative weighting on other driving actions included in the second presentation information Updating the driving behavior model;
A driving support method comprising:

(Item 2-7)
A predetermined number of driving actions are selected in descending order of the reliability of each of a plurality of types of driving actions, which are estimation results using the driving action model, and first presentation information indicating the selected predetermined number of driving actions is generated And steps to
Outputting the generated first presentation information to a notification device;
A step of inputting a first operation signal indicating one driving action selected by the occupant with respect to the first presentation information notified from the notification device;
Updating the driving behavior model by performing learning while performing positive weighting on one driving behavior indicated in the input first operation signal;
Generating a second presentation information including a driving action with low reliability instead of a part of the predetermined number of driving actions to be included in the first presentation information at a predetermined frequency instead of the first presentation information;
Outputting the generated second presentation information to the notification device;
A step of inputting a second operation signal indicating one driving action selected by the occupant with respect to the second presentation information notified from the notification device;
By performing positive weighting on one driving action indicated in the input second operation signal and performing learning while performing negative weighting on other driving actions included in the second presentation information A program for causing a computer to execute the step of updating the driving behavior model.

(Item 2-8)
A predetermined number of driving actions are selected in descending order of the reliability of each of a plurality of types of driving actions, which are estimation results using the driving action model, and first presentation information indicating the selected predetermined number of driving actions is generated A driving support device comprising: a generating unit that performs the presenting information output unit that outputs the first presentation information generated in the generating unit;
A notification device that notifies the presentation information output from the driving support device;
The driving support device includes:
An operation signal input unit that receives a first operation signal indicating one driving action selected by the occupant with respect to the first presentation information notified from the notification device;
A learning unit that updates the driving behavior model by performing learning while performing positive weighting on one driving behavior indicated in the first operation signal input to the operation signal input unit;
The generating unit, instead of a part of the predetermined number of driving actions to be included in the first presentation information, substitutes the second presentation information including driving actions with low reliability in place of the first presentation information at a predetermined frequency. Generate and
The presentation information output unit outputs the second presentation information generated by the generation unit to the notification device,
The operation signal input unit receives a second operation signal indicating one driving action selected by the occupant with respect to the second presentation information notified from the notification device,
The learning unit performs positive weighting on one driving action indicated in the second operation signal input to the operation signal input unit, and is negative with respect to other driving actions included in the second presentation information. A presentation system that updates a driving behavior model by performing learning while performing weighting.

  The present invention has been described based on the embodiments. It is understood by those skilled in the art that these embodiments are exemplifications, and that various modifications can be made to combinations of the respective constituent elements and processing processes, and such modifications are also within the scope of the present invention. By the way.

  In the first and second embodiments, the driving behavior estimation unit 70 is included in the control unit 41 of the driving support device 40. However, the present invention is not limited thereto, and for example, the driving behavior estimation unit 70 may be included in the control unit 31 of the automatic driving control device 30. According to this modification, the degree of freedom of configuration can be improved.

  In the first and second embodiments, the driving behavior model 80 is generated by the driving behavior learning unit 310 and transmitted to the driving behavior estimation unit 70. However, the present invention is not limited to this. For example, the driving behavior model 80 may be preinstalled in the driving behavior estimation unit 70. According to this modification, the configuration can be simplified.

  In the first and second embodiments, the driving behavior learning unit 310 may be included in the driving support device 40.

  In the first and second embodiments, the driving behavior estimation unit 70 uses a driving behavior model generated by deep learning using a neural network as an estimation. However, the present invention is not limited thereto, and for example, the driving behavior estimation unit 70 may use a driving behavior model using machine learning other than deep learning. An example of machine learning other than deep learning is SVM. Further, the driving behavior estimation unit 70 may use a filter generated by statistical processing. An example of a filter is collaborative filtering. In collaborative filtering, a driving action having a high correlation value is selected by calculating a correlation value between a driving history or a driving history corresponding to each driving action and test data. Since the certainty is indicated by the correlation value, the correlation value is also a likelihood and corresponds to the reliability. The learning unit 74 gives a reward for the correlation value as the reliability. According to this modification, the degree of freedom of configuration can be improved.

  In the first embodiment, the learning unit 74 gives a negative reward to the driving action that is not selected in the manual selection state. However, the present invention is not limited to this. For example, when the learning unit 74 is in the manual selection state, the negative reward may not be given to the driving action that is not selected. According to this modification, an increase in the difference between the reliability of the selected driving action and the reliability of the driving action that has not been selected can be suppressed.

  The combination of the first and second embodiments is also effective. According to this modification, the effect of the combination of the first and second embodiments can be obtained.

  2 Informing device, 2a Head-up display, 2b Center display, 4 Input device, 4a 1st operation part, 4b 2nd operation part, 4c 3rd operation part, 6 Speaker, 8 Radio | wireless apparatus, 10 Driving operation part, 11 Steering, 12 brake pedal, 13 accelerator pedal, 14 blinker switch, 20 detection unit, 21 position information acquisition unit, 22 sensor, 23 speed information acquisition unit, 24 map information acquisition unit, 30 automatic operation control device, 31 control unit, 32 storage unit , 33 I / O unit, 40 driving support device, 41 control unit, 42 storage unit, 43 I / O unit, 50 operation signal input unit, 51 image / sound output unit, 52 detection information input unit, 53 command IF, 54 Action information input part, 55 Command output part, 56 Communication I 70 driving behavior estimation unit, 72 display control unit, 74 learning unit, 80 driving behavior model, 82 estimation unit, 84 histogram generation unit, 90 generation unit, 92 processing unit, 94 selection unit, 100 vehicle, 300 server, 302 network 310 Driving behavior learning unit, 500 Driving support system.

Claims (8)

A generation unit that generates presentation information indicating a plurality of types of driving behavior, which is an estimation result using the driving behavior model;
A presentation information output unit that outputs the presentation information generated in the generation unit to a notification device;
An operation signal input unit that receives an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device;
When an operation signal is not input to the operation signal input unit, a selection unit that selects one driving action among a plurality of types of driving actions;
When an operation signal is input to the operation signal input unit, the driving behavior model is updated by performing learning while weighting one driving behavior indicated in the operation signal, and the operation signal input unit A learning unit that updates a driving behavior model by performing learning while weighting one driving behavior selected in the selection unit when an operation signal is not input;
The learning unit increases a weighting value when an operation signal is input to the operation signal input unit, than a weighting value when an operation signal is not input to the operation signal input unit. Driving assistance device.   When the operation signal is input to the operation signal input unit, the learning unit performs learning while performing negative weighting on a driving action other than one driving action indicated in the operation signal. The driving support device according to claim 1. The generation unit generates presentation information indicating some driving behaviors among a plurality of types of driving behaviors,
The learning unit, when an operation signal is input to the operation signal input unit, out of a part of driving behavior included in the presentation information generated in the generation unit, one driving behavior indicated in the operation signal The driving support device according to claim 2, wherein negative weighting is performed on driving behavior other than the above. A generation unit that generates presentation information indicating a plurality of types of driving behavior, which is an estimation result using the driving behavior model;
A presentation information output unit that outputs the presentation information generated in the generation unit to a notification device;
An operation signal input unit that receives an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device;
When an operation signal is not input to the operation signal input unit, a selection unit that selects one driving action among a plurality of types of driving actions;
When an operation signal is input to the operation signal input unit, automatic driving of the vehicle is controlled based on one driving action indicated in the operation signal, and no operation signal is input to the operation signal input unit An automatic driving control unit that controls automatic driving of the vehicle based on one driving action selected by the selection unit;
When an operation signal is input to the operation signal input unit, the driving behavior model is updated by performing learning while weighting one driving behavior indicated in the operation signal, and the operation signal input unit A learning unit that updates a driving behavior model by performing learning while weighting one driving behavior selected in the selection unit when an operation signal is not input;
The learning unit increases a weighting value when an operation signal is input to the operation signal input unit, than a weighting value when an operation signal is not input to the operation signal input unit. Automatic operation control device. A vehicle equipped with a driving support device,
The driving support device includes:
A generation unit that generates presentation information indicating a plurality of types of driving behavior, which is an estimation result using the driving behavior model;
A presentation information output unit that outputs the presentation information generated in the generation unit to a notification device;
An operation signal input unit that receives an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device;
When an operation signal is not input to the operation signal input unit, a selection unit that selects one driving action among a plurality of types of driving actions;
When an operation signal is input to the operation signal input unit, the driving behavior model is updated by performing learning while weighting one driving behavior indicated in the operation signal, and the operation signal input unit A learning unit that updates a driving behavior model by performing learning while weighting one driving behavior selected in the selection unit when an operation signal is not input;
The learning unit increases a weighting value when an operation signal is input to the operation signal input unit, than a weighting value when an operation signal is not input to the operation signal input unit. Vehicle. Generating presentation information indicating a plurality of types of driving behavior that is an estimation result using the driving behavior model;
Outputting the generated presentation information to a notification device;
An operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device; and
When an operation signal is not input, a step of selecting one driving action among a plurality of types of driving actions;
When an operation signal is input, updating the driving behavior model by performing learning while weighting one driving behavior indicated in the operation signal; and
A step of updating the driving behavior model by performing learning while weighting the one driving behavior selected in the selecting step when the operation signal is not inputted,
A driving support method, wherein a weighting value when an operation signal is input is made larger than a weighting value when an operation signal is not input. Generating presentation information indicating a plurality of types of driving behavior that is an estimation result using the driving behavior model;
Outputting the generated presentation information to a notification device;
An operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device; and
When an operation signal is not input, a step of selecting one driving action among a plurality of types of driving actions;
When an operation signal is input, updating the driving behavior model by performing learning while weighting one driving behavior indicated in the operation signal; and
A step of updating the driving behavior model by performing learning while weighting the one driving behavior selected in the selecting step when the operation signal is not inputted,
A program for causing a computer to execute a weighting value when an operation signal is input larger than a weighting value when an operation signal is not input. A driving support apparatus comprising: a generation unit that generates presentation information indicating a plurality of types of driving behaviors that are estimation results using a driving behavior model; and a presentation information output unit that outputs the presentation information generated in the generation unit; ,
A notification device that notifies the presentation information output from the driving support device;
The driving support device includes:
An operation signal input unit that receives an operation signal indicating one driving action selected by the occupant with respect to the presentation information notified from the notification device;
When an operation signal is not input to the operation signal input unit, a selection unit that selects one driving action among a plurality of types of driving actions;
When an operation signal is input to the operation signal input unit, the driving behavior model is updated by performing learning while weighting one driving behavior indicated in the operation signal, and the operation signal input unit A learning unit that updates the driving behavior model by performing learning while weighting one driving behavior selected by the selection unit when the operation signal is not input;
The learning unit increases a weighting value when an operation signal is input to the operation signal input unit, than a weighting value when an operation signal is not input to the operation signal input unit. To present system.

Images