JP7405050B2 - Automatic driving device, automatic driving method and program - Google Patents

Description

The present invention relates to an automatic driving device.

Conventionally, in the field of automated driving, attempts have been made to determine the driving behavior of a vehicle using the risk potential method. In the risk potential method, a cost function is evaluated based on the cumulative value of potential at each time within a predicted future time, and driving behavior is determined based on the evaluation results.

Japanese Patent Application Publication No. 2010-18062

When a vehicle drives automatically, it uses auto lane changing (ALC), cruise control (CC), adaptive cruise control (ACC), and lane keeping system (LKS). Various functions such as collision damage mitigation braking (AEB: Autonomous Emergency Braking), automatic emergency steering (AES: Automatic Emergency Steering), etc. are optimally combined and operated according to the surrounding situation. If you try to set such combinations of multiple functions according to the traffic conditions around your vehicle, you will need to design transition conditions and operations for a huge number of combinations at the development stage, which will require a large amount of development man-hours. The problem arose that it required Another problem arises in that it is insufficient in its versatility or robustness in complex traffic environments such as urban areas or highway junctions.

Therefore, the present invention has been made in view of these points, and an object of the present invention is to improve versatility or robustness during automatic driving in a complicated traffic environment.

An automatic driving device according to one aspect of the present invention is an automatic driving device that determines the driving behavior of its own vehicle based on a potential field, and the automatic driving device determines the driving behavior of its own vehicle based on a potential field. long-term prediction that selects a driving mode in which the own vehicle will travel based on a long-term collision risk when traveling on the travel route until the predicted time and a distance that can be traveled until the long-term predicted time; and a short-term collision risk based on the short-term collision risk between the current time and a short-term predicted time shorter than the long-term predicted time while the host vehicle is traveling based on the driving mode selected by the long-term prediction unit. a short-term prediction unit that determines driving behavior; and a short-term prediction unit that determines the short-term driving behavior based on the ultra-short-term collision risk between the current time and an ultra-short-term prediction time shorter than the short-term prediction time when the short-term collision risk is equal to or greater than a threshold value. and an ultra-short-term prediction unit that determines different ultra-short-term driving behavior.

The long-term prediction unit may calculate the long-term collision risk on the assumption that the obstacle detected by the host vehicle continues to move over the long-term prediction time.

The short-term prediction unit determines whether the obstacle detected by the own vehicle moves during the short-term prediction time, remains stationary during the short-term prediction time, or moves only for a part of the short-term prediction time. The short-term driving behavior may be determined based on the short-term collision risk calculated for the short-term collision risk.

The short-term prediction unit may determine the short-term driving behavior based on the short-term collision risk and a destination potential and a lane potential corresponding to the driving mode selected by the long-term prediction unit.

The very short-term prediction unit calculates the very short-term collision risk when the obstacle detected by the own vehicle moves over the very short-term prediction time or remains stationary over the very short-term prediction time, and the long-term prediction unit The very short-term driving behavior may be determined based on the lane potential corresponding to the selected driving mode.

The long-term prediction unit, the short-term prediction unit, and the very short-term prediction unit perform driving behavior on the assumption that the obstacle detected by the host vehicle moves according to both the speed and acceleration at the time the obstacle was detected. You may decide.

According to the present invention, it is possible to determine driving behavior with excellent versatility and robustness during automatic driving in a complex traffic environment.

It is a schematic diagram for explaining an example of a composition of an automatic driving device. FIG. 3 is a diagram showing an overview of multiple predictions. FIG. 3 is a diagram showing an overview of multiple predictions. FIG. 3 is a diagram showing the configuration of a control device. It is a flowchart which shows the flow of processing in a control device. It is a flowchart showing the flow of short-term prediction. It is a flowchart showing the flow of ultra-short-term prediction.

[Overview of automatic driving device]
An overview of an automatic driving device according to one embodiment of the present invention will be described with reference to FIG. 1. The automatic driving device according to the present embodiment is, for example, a device that is installed in a vehicle and executes various processes for automatically driving the vehicle.

FIG. 1 is a schematic diagram for explaining an example of the configuration of an automatic driving device 1. As shown in FIG. The automatic driving device 1 is mounted on a vehicle such as a truck, and supports driving of the own vehicle. The automatic driving device 1 determines a driving mode and driving behavior during automatic driving, for example. The driving mode is determined by a combination of a driving route starting from the current position of the own vehicle and a target driving speed. The driving behavior is the movement of the own vehicle starting from the current time, which is determined by, for example, the acceleration/deceleration and turning angular velocity (hereinafter referred to as "yaw rate") of the own vehicle. The automatic driving device 1 determines the driving mode and driving behavior of the own vehicle based on the potential field.

As shown in FIG. 1, the automatic driving device 1 includes a vehicle detection section 2, an environment recognition section 4, a map database 6, and a control device 10.
Vehicle detection unit 2 detects the state of the own vehicle. The vehicle detection unit 2 detects the position and speed of the own vehicle. For example, the vehicle detection unit 2 includes a GPS (Global Positioning System) receiver, and detects the position of the own vehicle using radio waves received by the GPS receiver. The vehicle detection unit 2 outputs traveling information indicating the position and speed of the own vehicle based on the detection result of the own vehicle position to the control device 10.

The environment recognition unit 4 recognizes the environmental situation around the host vehicle. For example, the environment recognition unit 4 includes an external sensor such as a camera or radar. The environment recognition unit 4 recognizes objects around the host vehicle (eg, other vehicles, bicycles, pedestrians, etc.) based on outputs from external sensors. Furthermore, the environment recognition unit 4 can recognize, for example, the position and width of the lane in which the host vehicle is traveling, as well as the travelable area. The environment recognition unit 4 outputs surrounding information indicating the recognition result of the surrounding environment to the control device 10.

The map database 6 stores road map information. The road map information includes, for example, data indicating three-dimensional coordinates of latitude, longitude, and altitude of a road. Further, the road map information includes information on the number of lanes and the lane structure of the road on which the vehicle is traveling. Furthermore, the map database 6 can instead acquire information on lanes recognized by the environment recognition unit 4 based on the position of the own vehicle detected by the vehicle detection unit 2.

The control device 10 controls the operation of the automatic driving device 1. The control device 10 determines the driving behavior of the own vehicle using the potential field. The potential field is, for example, a destination potential, an obstacle potential or a lane potential. The destination potential is a potential based on attraction or repulsion toward a preset destination. The obstacle potential is a potential based on a repulsive force from an obstacle. The lane potential is the potential to follow a predetermined lane.

Although details will be described later, the control device 10 of the present embodiment determines driving behavior that can minimize a cost function that includes the cumulative value of potential over a predetermined period and the maximum collision risk as an actual risk. . Thereby, the automatic driving device 1 can prevent driving behavior from being determined depending on the size of an obstacle, and can realize safer automatic driving.

The control device 10 uses long-term predictions, short-term predictions, and very short-term predictions that have different prediction times and prediction methods, respectively, in order to determine driving behavior. FIGS. 2 and 3 are diagrams showing an overview of these multiple predictions.

Long-term prediction is a process of determining which driving mode is best to select from among a plurality of driving modes by calculating the collision risk over the longest period of time, and is, for example, an automatic lane change process. FIG. 3(a) shows a plurality of driving modes that are candidates to be selected in the long-term prediction. In long-term prediction, for example, a driving mode I ₀ in which the vehicle continues to drive in the lane in which the vehicle is traveling, a driving mode I R in which the vehicle changes to the right lane of the lane in which the vehicle is traveling, and a driving mode I _R in which the vehicle changes to the lane on the right side of the lane in which the vehicle is traveling. Among the driving modes _IL in which the driver changes lanes to the left lane, a driving mode is selected in which the risk of collision is relatively small and the distance traveled within the time period that is subject to long-term prediction is relatively large.

The long-term prediction is based on the driving mode based on a long-term cost function whose parameters are the collision risk over Δt _L = 10 seconds (hereinafter sometimes referred to as "long-term collision risk") and the distance that can be traveled during Δt _L. This is a process of selecting a short-term prediction, which will be described later, and is executed at a cycle of about 200 milliseconds to 1 second, for example, because it involves a large calculation load compared to short-term prediction.

式（１）における第１項は、各運転モードに対応する経路における無次元化された最終到達距離（Δｔ_Ｌの間に走行可能な距離）であり、第２項は、各経路を通過する際の正規化された最大リスクである。式（１）はこれらの２項の重み付き和として長期コスト関数を定義している。ここで、V^targetは目標速度、S_tendは経路の終点までの道のり距離、R_max ^collは経路中の最大衝突リスク、R_th ^collは経路生成を打ち切るリスクの閾値である。 The long-term cost function used in long-term prediction is expressed, for example, by the following equation (1).

The first term in Equation (1) is the nondimensional final reach distance (distance that can be traveled during Δt _L ) on the route corresponding to each driving mode, and the second term is the distance traveled through each route. is the normalized maximum risk. Equation (1) defines a long-term cost function as a weighted sum of these two terms. Here, V ^target is the target speed, S _tend is the distance to the end of the route, R _max ^coll is the maximum collision risk on the route, and R _th ^coll is the risk threshold for aborting route generation.

Since the long-term prediction time is relatively long, it is unlikely that a moving dynamic obstacle will remain in the same location. In addition, the long-term collision risk detected in long-term prediction can be avoided by changing driving behavior. The cost function is calculated assuming that the object continues to move at the same acceleration/deceleration rate.

Short-term prediction is a prediction of near-future conditions based on a potential field corresponding to the driving mode determined in long-term prediction, and for example, driving behavior corresponding to a lane keeping system, inter-vehicle distance control device, or vehicle speed maintenance system is executed. . Short-term prediction is a process of selecting driving behavior by predicting a collision risk in a short-term prediction time that is shorter than a long-term prediction time. Short-term prediction involves determining driving behavior based on a short-term cost function whose parameters are, for example, the collision risk for Δt _S = 2 seconds to 5 seconds (hereinafter sometimes referred to as "short-term collision risk"), destination potential, and lane potential. This is a selection process. Short-term prediction is executed at a cycle of, for example, about 10 to 50 milliseconds because it is necessary to determine the most recent driving behavior.

FIG. 3B shows a plurality of driving behaviors that are candidates to be selected in the short-term prediction using a plurality of arrows. In short-term prediction, the driving behavior that minimizes the short-term cost function is selected from among these multiple driving behaviors.

式（２）における第１項は基本走行ポテンシャルU^baseの短期予測時間内における累積値であり、第２項は複数の運転行動それぞれの衝突リスクのうち最大のリスクR_max ^collと予測時間の積である。また、(X_t ^front, Y_t ^front)は自車両前面位置の座標を表す。なお、基本走行ポテンシャルは目的地ポテンシャル及び車線ポテンシャルを含んでいる。U^dst(X,Y)は目的地ポテンシャルであり、U^laneは車線ポテンシャルである。これらを加算した値が式（２）における基本走行ポテンシャルU^baseである。 The short-term cost function is expressed, for example, by the following equation (2).

The first term in equation (2) is the cumulative value of the basic driving potential U ^base within the short-term prediction time, and the second term is the product of the maximum risk R _max ^coll of the collision risks of each of multiple driving actions and the prediction time. It is. Furthermore, (X _t ^front , Y _t ^front ) represents the coordinates of the front position of the own vehicle. Note that the basic driving potential includes a destination potential and a lane potential. U ^dst (X,Y) is the destination potential, and U ^lane is the lane potential. The value obtained by adding these is the basic running potential U ^base in equation (2).

The basic driving potential includes a destination potential for directing the vehicle to a preset destination at a set speed (for example, a speed limit) and a lane potential for causing the vehicle to follow the lane. The destination potential is, for example, a potential that is the sum of a speed difference potential that is proportional to the speed difference between the target speed and the own vehicle's speed, and an azimuth difference potential that is proportional to the azimuth difference between the target position and the own vehicle's position. be. Further, the lane potential includes, for example, an attractive force potential that causes the own vehicle to be generated in the center of a predetermined lane, and a repulsive force potential that is generated at both ends of the lane.

In short-term forecasting, it is important to prevent the own vehicle from putting itself in a dangerous situation. Therefore, in short-term prediction, driving behavior is determined in consideration of the risks in both cases, for example, when a dynamic obstacle moves and when a dynamic obstacle remains in the same place.

Ultra-short-term prediction is a process that selects driving behavior by predicting the collision risk in an ultra-short-term prediction time that is shorter than the short-term prediction time. Equivalent to. Ultra-short-term prediction is a process of selecting driving behavior based on an ultra-short-term cost function whose parameter is a collision risk over a period of about Δt _S' = 1 second (hereinafter sometimes referred to as "ultra-short-term collision risk"). be.

FIG. 3(c) shows a plurality of driving behaviors that are candidates to be selected in the ultra-short-term prediction with a plurality of arrows. In the ultra-short-term prediction, the driving behavior that minimizes the ultra-short-term cost function is selected from among these multiple driving behaviors.

超短期予測においては、危険が目前に迫っている状況なので、動的障害物の位置が超短期予測時間にわたって移動するという前提で、衝突を回避可能な安全な領域を探索する。超短期予測においては、障害物を緊急回避することが重要なので、運転行動を探索する際の自車両の加減速度及び旋回角速度の許容値が短期予測よりも大きくする。 The ultra-short-term cost function is expressed, for example, by the following equation (3).

In ultra-short-term prediction, since danger is imminent, a safe area where a collision can be avoided is searched on the premise that the position of the dynamic obstacle will move over the ultra-short-term prediction time. In ultra-short-term prediction, since it is important to urgently avoid obstacles, the allowable values for acceleration/deceleration and turning angular velocity of the own vehicle when searching for driving behavior are set larger than in short-term prediction.

As described above, the automatic driving device 1 determines driving behavior using long-term prediction, short-term prediction, and very short-term prediction for different purposes. Then, the automatic driving device 1 performs long-term prediction, short-term prediction, and very short-term prediction in parallel at different cycles. Short-term predictions and very short-term predictions are performed in the same cycle, and long-term predictions are performed in a longer cycle than the short-term predictions and very short-term predictions. By operating the automatic driving device 1 in this way, the delay time until driving behavior is determined can be reduced compared to the case where short-term prediction and ultra-short-term prediction are performed after long-term prediction is completed. , driving behavior during automated driving can be determined in real time.
The configuration and operation of the control device 10 will be described in detail below.

[Detailed configuration of control device 10]
FIG. 4 is a diagram showing the configuration of the control device 10. The control device 10 includes a storage section 12 and a control section 14. The storage unit 12 includes, for example, a ROM (Read Only Memory) and a RAM (Random Access Memory). The storage unit 12 stores programs and various data for the control unit 14 to execute.

The control unit 14 is, for example, a CPU (Central Processing Unit). The control unit 14 functions as an information acquisition unit 141, a long-term prediction unit 142, a short-term prediction unit 143, a very short-term prediction unit 144, and a driving behavior determination unit 145 by executing the program stored in the storage unit 12.

The information acquisition unit 141 acquires driving information of the own vehicle. For example, the information acquisition unit 141 acquires the position and speed of the own vehicle while it is running. The information acquisition unit 141 acquires the position and speed of the own vehicle from the detection results of the vehicle detection unit 2 (FIG. 1).

The information acquisition unit 141 also acquires information regarding objects around the own vehicle. For example, the information acquisition unit 141 acquires surrounding information regarding objects (other vehicles, pedestrians, etc.) existing in the traveling direction of the host vehicle from the detection results of the environment recognition unit 4 (FIG. 1).

The information acquisition unit 141 also acquires target information regarding the target position and target speed at which the vehicle should travel. The target position is, for example, a position a predetermined distance (for example, 200 m) in front of the host vehicle. The target position may be an arbitrary point set in the road map information stored in the map database 6. The target speed is, for example, the legal speed for the lane. The information acquisition unit 141 can, for example, refer to road map information stored in the map database 6 to acquire target information. The information acquisition unit 141 inputs the acquired driving information, surrounding information, and road map information to the long-term prediction unit 142, the short-term prediction unit 143, and the very short-term prediction unit 144.

長期予測部１４２は、移行した運転モードを運転行動決定部１４５に通知する。長期予測部１４２は、移行した運転モードを短期予測部１４３に通知してもよい。 [Long-term forecast]
The long-term prediction unit 142 executes long-term prediction. The long-term prediction unit 142 calculates the long-term collision risk when traveling on the driving route between the current time and the long-term prediction time, and the long-term collision risk between the current time and the long-term prediction time, from a plurality of driving modes with different combinations of driving routes and target speeds. The driving mode in which the host vehicle travels is selected based on the travelable distance. The long-term prediction unit 142 calculates the long-term cost function C ^path (I) for each of the plurality of driving mode candidates using, for example, the above equation (1), and calculates the long-term cost function C path (I) based on the following equation (4). Select the operating mode that minimizes the cost function. Furthermore, the long-term prediction unit 142 determines whether the own vehicle is running stably, whether the blinker is turned on, and shifts to the selected driving mode.

The long-term prediction unit 142 notifies the driving behavior determining unit 145 of the transitioned driving mode. The long-term prediction unit 142 may notify the short-term prediction unit 143 of the shifted driving mode.

Note that the long-term prediction unit 142 may calculate the long-term collision risk on the assumption that the obstacle detected by the host vehicle continues to move over the long-term prediction time. For example, the long-term prediction unit 142 may calculate the long-term collision risk on the assumption that the obstacle continues to move at the speed and acceleration at the time of detection.

[Short-term forecast]
The short-term prediction unit 143 calculates a short-term collision risk based on the short-term collision risk between the current time and a short-term prediction time shorter than the long-term prediction time while the own vehicle is traveling based on the driving mode selected by the long-term prediction unit 142. Determine driving behavior. The short-term prediction unit 143 acquires the driving mode selected by the long-term prediction unit 142, and determines the optimal short-term driving behavior in the potential field in the acquired driving mode. That is, the short-term prediction unit 143 determines short-term driving behavior based on the short-term collision risk and the destination potential and lane potential corresponding to the driving mode selected by the long-term prediction unit 142.

The short-term prediction unit 143 determines short-term driving behavior by updating parameters related to the potential field based on the driving mode determined by the long-term prediction unit 142 and calculating a short-term cost function. The short-term prediction unit 143 notifies the driving behavior determining unit 145 of the determined short-term driving behavior. The short-term prediction unit 143 may notify the very short-term prediction unit 144 of the determined short-term driving behavior.

The short-term prediction unit 143 calculates, for each of a plurality of driving actions, a short-term cost function that includes as a parameter the short-term collision risk during driving between the current time and the short-term prediction time. The short-term prediction unit 143 can calculate a short-term cost function for each of a plurality of driving behavior candidates based on the above equation (2), for example. A short-term cost function may be calculated that includes the potential and lane potential as parameters. Further, the short-term prediction unit 143 may determine the acceleration/deceleration and yaw rate of the own vehicle as the driving behavior of the own vehicle. Then, the short-term prediction unit 143 may calculate a short-term cost function having a regularization term for driving behavior that includes the acceleration/deceleration and yaw rate of the own vehicle as parameters, as shown in Equation (5) below.

The short-term prediction unit 143 uses the information acquired by the information acquisition unit 141 to identify the destination potential and lane potential, thereby identifying the basic driving potential.

The third and fourth terms in equation (5) are regularization terms necessary for smooth operation. α means acceleration/deceleration, γ means yaw rate, w _α is the weight of acceleration/deceleration, and w _γ is the weight of yaw rate.

The short-term prediction unit 143 calculates the risk of collision with objects around the own vehicle within a predetermined prediction time as the actual risk included in the short-term cost function. For example, the short-term prediction unit 143 calculates the maximum collision risk for the object detected by the environment recognition unit 4. If there are multiple objects in the vicinity, the short-term prediction unit 143 calculates the maximum collision risk for each object. The short-term prediction unit 143 determines the maximum collision risk from among the determined maximum collision risks for the plurality of objects.

Using maximum collision risk has the following advantages: Here, it is assumed that there are large objects (e.g., other vehicles) and small objects (e.g., pedestrians) in front of the vehicle in the direction of travel, and it is necessary to avoid the small objects for safety. . When the cumulative value of collision risk is used, driving behavior depends on the size of the object, so there is a risk that the driver will avoid the large object and head toward the small object. On the other hand, using maximum collision risk allows safe driving behavior by avoiding small, high-risk objects.

The short-term prediction unit 143 determines the optimal driving behavior (acceleration/deceleration and yaw rate) that minimizes the short-term cost function based on the following equation (6).

In this way, when the short-term prediction unit 143 selects the driving behavior that minimizes the value of the short-term predicted cost function, it is assumed that an obstacle in front of the vehicle or an obstacle in the rear that is approaching the own vehicle exists in a very close position. In this case, no matter which driving behavior (acceleration/deceleration and yaw rate) the short-term prediction unit 143 selects, the collision risk may end up being a high value. If the own vehicle performs such driving behavior, priority will be given to maintaining speed or lane maintenance, and there is a risk that the vehicle will collide with an obstacle.

Therefore, the short-term prediction unit 143 compares the short-term collision risk when the determined optimal driving behavior is performed with the threshold value _Rth , and determines whether it is okay to drive the own vehicle according to the optimal driving behavior, or whether the optimal driving behavior is dangerous. It is determined whether other driving actions are necessary based on very short-term predictions. The threshold value R _th is determined, for example _, based on the likelihood of sudden acceleration/deceleration (for example, sudden braking). becomes more likely to be selected. When the short-term prediction unit 143 instructs the ultra-short-term prediction unit 144 to determine a driving behavior, the short-term prediction unit 143 generates at least an audible or visual warning indicating that there is a risk of a collision or that sudden acceleration/deceleration will be performed. Depending on the situation, the occupants of the own vehicle may be notified.

In order to improve safety, the short-term prediction unit 143 determines whether an obstacle detected by the own vehicle moves over the short-term prediction time, remains stationary over the short-term prediction time, or for a part of the short-term prediction time. The short-term driving behavior may be determined based on the short-term collision risk calculated for the case where the vehicle only moves. The short-term prediction unit 143 may determine short-term driving behavior based on short-term collision risks calculated for a plurality (for example, all) of these cases. For example, the short-term prediction unit 143 selects a short-term driving behavior in which the short-term collision risk is less than the threshold value R _th from among the short-term collision risks based on all the calculated driving behaviors, and applies the selected short-term driving behavior to the driving behavior determining unit. 145. If a plurality of short-term collision risks are less than the threshold value R _th , the short-term prediction unit 143 selects the short-term driving behavior that minimizes the short-term collision risk.

On the other hand, if all the calculated short-term collision risks are equal to or greater than the threshold, or if the short-term collision risk when the short-term cost function is minimized is equal to or greater than the threshold _Rth , the short-term prediction unit 143 calculates The very short-term prediction unit 144 is instructed to determine the driving behavior based on the following information.

[Very short-term forecast]
The ultra-short-term prediction unit 144 determines ultra-short-term driving behavior that is different from the short-term driving behavior based on the ultra-short-term collision risk between the current time and the ultra-short-term prediction time that is shorter than the short-term prediction time, when the short-term collision risk is equal to or higher than the threshold value. Determine. For example, if all the short-term collision risks calculated by the short-term prediction unit 143 are equal to or higher than the threshold, or if the short-term collision risk when the short-term cost function is minimized is equal to or higher than the threshold, the very short-term prediction unit 144 calculates the The ultra-short-term collision risk during driving up to the ultra-short-term prediction time, which is shorter than the short-term prediction time, and the ultra-short-term cost function that includes the ultra-short-term collision risk as a parameter are calculated. The very short-term prediction unit 144 may calculate a very short-term cost function that includes a very short-term collision risk, destination potential, and lane potential as parameters.

The ultra-short-term prediction unit 144 can calculate an ultra-short-term cost function that includes ultra-short-term collision risk and lane potential as parameters for each of the plurality of driving behavior candidates, for example, based on the above equation (3). , the very short-term prediction unit 144 determines driving behavior using the maximum risk R _max ^coll in the very short-term prediction as a long-term and short-term cost function, as shown by the following equation (7). Alternatively, as shown in equations (8), (9), and (10), a very short-term cost function that considers only the lane potential as the basic travel potential, a very short-term cost function that considers both the destination potential and the lane potential, and A very short-term cost function may be calculated that has a regularization term that includes the acceleration/deceleration and yaw rate of the host vehicle as parameters.

The acceleration/deceleration weight w' _α and the yaw rate weight w' _α in the third and fourth terms of Equation (10) are the third and fourth terms of the short-term cost prediction function (Equation (5)) in short-term prediction. It is smaller than the acceleration/deceleration weight w _α and the yaw rate weight w _γ included in . That is, the influence of acceleration/deceleration and yaw rate on the very short-term cost function is smaller than the influence of acceleration/deceleration and yaw rate on the short-term cost function, and the very short-term prediction unit 144 tends to increase the acceleration/deceleration and yaw rate. In this case, the short-term prediction unit 143 calculates a short-term cost function that includes a regularization term for driving behavior of the own vehicle, and the very short-term prediction unit 144 includes a regularization term for driving behavior that has a smaller weight than the short-term cost function. Calculate the very short-term cost function. In this way, the very short-term prediction unit 144 may calculate a very short-term cost function that includes acceleration/deceleration and yaw rate, which have smaller weights than the short-term cost function, as parameters. When the ultra-short-term prediction unit 144 uses such an ultra-short-term cost function, it becomes possible to avoid danger by suddenly applying the brakes or accelerating.

The very short-term prediction unit 144 determines the optimal driving behavior (acceleration/deceleration and yaw rate) that minimizes the very short-term cost function based on the following equation (11). The very short-term prediction unit 144 replaces the optimal driving behavior determined by the short-term prediction unit 143 with the optimal driving behavior determined based on the very short-term cost function.

The very short-term prediction unit 144 calculates the very short-term collision risk when the obstacle detected by the own vehicle moves over the very short-term predicted time or remains stationary over the very short-term predicted time, and the driving mode selected by the long-term predicted unit. Very short-term driving behavior may be determined based on the corresponding lane potential. By operating in this way, the ultra-short-term prediction unit 144 selects safe driving behavior in situations where danger is imminent, on the premise that the position of the dynamic obstacle will continue to move during the ultra-short-term prediction time. can do.

ここで、d_minは周辺の物体との最小の距離であり、d_aveは複数の周辺物体との平均距離である。 If the surrounding obstacles cannot be avoided by minimizing the risk, the very short-term prediction unit 144 may calculate a very short-term cost function that includes the distance from the host vehicle to the surrounding obstacles as a parameter. That is, the very short-term prediction unit 144 may determine the driving behavior based on a very short-term cost function that includes the distance to surrounding objects as a parameter. The ultra-short-term cost function introduced as a parameter of the distance to the obstacle to equation (7) is expressed by the following equation (12).

Here, d _min is the minimum distance to surrounding objects, and d _ave is the average distance to a plurality of surrounding objects.

d _min and d _ave are calculated by the following equation (13). (X _ts' ^ego , Y _ts' ^ego ) indicates the position of the own vehicle after time t _s' , and (X _i,ts' ^obj , Y _i,ts' ^obj ) indicates the i-th position after time t _s' . Indicates the position of an object.

The very short-term prediction unit 144 determines the driving behavior based on the collision risk in the very short-term prediction time, which is shorter than the short-term prediction time that is the time for the short-term prediction unit 143 to select the driving behavior. Because the ultra-short-term prediction unit 144 makes predictions in such a short time, even if the short-term collision risk corresponding to the driving behavior selected by the short-term prediction unit 143 is large, the ultra-short-term prediction time is shorter than the short-term prediction time. It is easy to distinguish between relatively dangerous driving behavior and relatively safe driving behavior within the range of . Note that the long-term prediction unit 142, the short-term prediction unit 143, and the very short-term prediction unit 144 determine the driving behavior on the assumption that the obstacle moves according to both the speed and acceleration at the time when the obstacle is detected. Good too.

[Decision of driving behavior]
The driving behavior determining unit 145 determines to drive the own vehicle according to the driving behavior notified from the very short-term prediction unit 144 or the driving behavior notified from the driving behavior determining unit 145, and generates driving behavior data indicating the determined driving behavior. is notified to the device or unit that controls the driving of the own vehicle. That is, when the ultra-short-term prediction unit 144 calculates the ultra-short-term cost function, the driving behavior determining unit 145 determines the driving behavior based on the ultra-short-term cost function, and when the ultra-short-term prediction unit 144 calculates the ultra-short-term cost function. If not, determine driving behavior based on short-term cost functions.

[Flowchart of processing in control device 10]
FIG. 5 is a flowchart showing the flow of processing in the control device 10. The control device 10 repeats the processes from S11 to S18 shown in FIG. 5 while the engine of the host vehicle is started.

First, the information acquisition unit 141 acquires various information from the vehicle detection unit 2, environment recognition unit 4, and map database 6 (S11), and calculates a potential map based on the acquired information (S12). The potential map is data indicating destination potential and lane potential.

The long-term prediction unit 142 selects the optimal driving mode from a plurality of driving modes based on the potential map (S13). Next, the short-term prediction unit 143 selects the optimal driving behavior by executing short-term prediction using the potential corresponding to the driving mode selected by the long-term prediction unit 142 (S14). The flow of short-term prediction processing will be described later.

If the short-term collision risk corresponding to the optimal short-term driving behavior selected by the short-term prediction unit 143 is equal to or higher than the threshold (YES in S15), it is determined that a very short-term prediction is necessary, and the very short-term prediction unit 143 determines that a very short-term prediction is necessary. Instruct the short-term prediction unit 144. The very short-term prediction unit 144 executes very short-term prediction based on the instruction (S16). The flow of the ultra-short-term prediction process will be described later.

If the short-term collision risk corresponding to the optimal short-term driving behavior selected by the short-term prediction unit 143 is less than the threshold (NO in S15), the driving behavior determination unit 145 adjusts the short-term driving behavior selected by the short-term prediction unit 143 to the own vehicle. The driving behavior is determined (S17). If the short-term collision risk corresponding to the optimal short-term driving behavior selected by the short-term prediction unit 143 is greater than or equal to the threshold (YES in S15), the driving behavior determination unit 145 selects the ultra-short-term driving behavior selected by the ultra-short-term prediction unit 144. The driving behavior of the own vehicle is determined (S17). The control device 10 repeats the processes from S11 to S18 until the operation ends (NO in S18).

[Flowchart of short-term forecast]
FIG. 6 is a flowchart showing the flow of short-term prediction. First, the short-term prediction unit 143 selects, from a plurality of driving behaviors, a driving behavior for which a short-term cost function is calculated and suitability is investigated (S141). Next, the short-term prediction unit 143 identifies the state of the own vehicle at time _ts (S142). Specifically, the short-term prediction unit 143 identifies the position (for example, the X coordinate and Y coordinate of the center of the rear wheel axis), yaw angle, and speed of the own vehicle.

Next, the short-term prediction unit 143 calculates a basic driving potential (S143). Specifically, the short-term prediction unit 143 calculates the basic driving potential by adding the destination potential and the lane potential. Further, the short-term prediction unit 143 calculates the collision risk within the short-term prediction time (S144). The short-term prediction unit 143 calculates a cost based on a short-term cost function including the calculated basic driving potential and collision risk as parameters (S145), and stores the calculated cost in the storage unit 12 in association with driving behavior.

The short-term prediction unit 143 determines whether costs have been calculated for all selectable driving actions (S146), and if all driving actions have not been selected and costs have been calculated (NO in S146), S141 Return processing to . When the short-term prediction unit 143 calculates the cost for all selectable driving behaviors (YES in S146), the short-term prediction unit 143 selects the driving behavior with the minimum cost among the plurality of driving behaviors stored in the storage unit 12 as the short-term driving behavior. It is determined (S147).

[Flowchart of ultra-short-term prediction]
FIG. 7 is a flowchart showing the flow of very short-term prediction. First, the very short-term prediction unit 144 selects a driving behavior from among a plurality of driving behaviors that is to be investigated for suitability by calculating a very short-term cost function (S161). Next, the very short-term prediction unit 144 identifies the state of the own vehicle (S162). Specifically, the very short-term prediction unit 144 identifies the position, yaw angle, and speed of the host vehicle at time t _s' .

Next, the very short-term prediction unit 144 calculates lane potential (S163). Further, the very short-term prediction unit 144 calculates the collision risk within the short-term prediction time (S164). The ultra-short-term prediction unit 144 calculates a cost based on an ultra-short-term cost function that includes the calculated lane potential and collision risk as parameters (S165), and stores the calculated cost in the storage unit 12 in association with the driving behavior.

The ultra-short-term prediction unit 144 determines whether costs have been calculated for all selectable driving actions (S166), and if all driving actions have not been selected and costs have been calculated (NO in S166), The process returns to S161. When the cost has been calculated for all selectable driving actions (YES in S166), the very short-term prediction unit 144 calculates the driving action with the minimum cost among the plurality of driving actions stored in the storage unit 12 as the very short-term prediction unit 144. The driving action is determined (S167).

[Effects of control device 10]
As described above, the control device 10 includes the potential as a parameter and determines the final driving behavior by minimizing the respective cost functions of the different long-term predictions, short-term predictions, and very short-term predictions. By changing the parameters included in the cost function according to the purpose of each prediction, the control device 10 can determine the optimal driving behavior according to the traffic conditions around the host vehicle. As a result, various functions such as automatic lane change (ALC), vehicle speed maintenance system (CC), inter-vehicle distance control system (ACC), lane keeping system (LKS), collision mitigation braking (AEB), automatic steering avoidance (AES), etc. It can be operated optimally according to the surrounding circumstances of the driving behavior corresponding to the function. Therefore, it becomes possible to determine driving behavior with excellent versatility and robustness that not only ensures safety in complex traffic environments in automated driving but also follows driving norms, improving safety and ensuring standard driving. realizable.

If you try to set combinations of multiple driving support functions as described above in advance for each traffic situation around your vehicle, the operation of a huge number of combinations needs to be determined at the development stage, which requires a large amount of development man-hours. It takes. Furthermore, the versatility and robustness for complex traffic environments such as urban areas or expressway junctions are insufficient. On the other hand, like the control device 10 according to the present embodiment, the driving mode is determined based on long-term prediction, and short-term prediction and very short-term prediction are used based on the driving potential in the determined driving mode. By determining driving behavior, versatility and robustness can be improved.

In addition, the very short-term prediction unit 144 determines whether all the short-term collision risks calculated by the short-term prediction unit 143 are equal to or greater than the threshold, or if the short-term collision risk when the short-term cost function is minimized is equal to or greater than the threshold. The short-term collision risk is calculated, and driving behavior is determined based on a very short-term cost function that includes the very short-term crash risk as a parameter. In this way, when the driving behavior determined in the short-term prediction does not sufficiently reduce the risk, the control device 10 uses ultra-short-term prediction with relaxed conditions for acceleration/deceleration and yaw rate to take driving behavior that can avoid the risk. decide.

By operating the control device 10 in this manner, it becomes possible to avoid a collision risk that cannot be avoided when automatic driving is performed based on a single length of predicted time. Furthermore, it becomes easier to perform appropriate driving behavior to ensure maximum safety even in the event of a sudden jump or interruption.

Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments, and various modifications and changes can be made within the scope of the gist. be. For example, all or part of the device can be functionally or physically distributed and integrated into arbitrary units. In addition, new embodiments created by arbitrary combinations of multiple embodiments are also included in the embodiments of the present invention. The effects of the new embodiment resulting from the combination have the effects of the original embodiment.

1 Automatic driving device 2 Vehicle detection unit 4 Environment recognition unit 6 Map database 10 Control device 12 Storage unit 14 Control unit 141 Information acquisition unit 142 Long-term prediction unit 143 Short-term prediction unit 144 Very short-term prediction unit 145 Driving behavior determination unit

Claims (5)

Based on the destination potential corresponding to the strength of gravitational force toward a preset destination and the lane potential corresponding to the strength of gravitational force toward the center of the lane in which the vehicle is traveling, the current position of the vehicle is used as the starting point. An automatic driving device that determines a driving behavior that is a driving mode determined by a combination of a driving route and a target speed, and a movement of the own vehicle starting from a current point determined by the acceleration/deceleration and turning angular velocity of the own vehicle ,
A long-term collision risk indicating the magnitude of the risk of the own vehicle colliding with an obstacle when traveling on the driving route during the long-term prediction period from the current time to the long-term prediction time, from the plurality of driving modes , and the long-term prediction period . a long-term prediction unit that predicts the driving mode that is optimal for the host vehicle to travel during the long-term prediction period so that a long-term cost function represented by
While the host vehicle is traveling based on the driving mode predicted by the long-term prediction unit, the host vehicle collides with an obstacle during a short-term prediction period from the current time to a short-term prediction time shorter than the long-term prediction time. When the host vehicle travels during the short-term prediction period, the vehicle is configured to a short-term prediction unit that predicts short-term driving behavior that is the optimal driving behavior ;
If the short-term collision risk when the short-term cost function is minimized is greater than or equal to a threshold, the own vehicle collides with an obstacle during a very short-term prediction period from the present to a very short-term prediction time shorter than the short-term prediction time. The driving behavior that is different from the short-term driving behavior is set during the very short-term prediction period so that the very short-term collision risk indicating the size of the risk of the collision and the very short-term cost function expressed by the lane potential are minimized. a very short-term prediction unit that predicts the optimal driving behavior when the vehicle is running ;
An automatic driving device equipped with. The long-term prediction unit calculates the long-term collision risk assuming that the obstacle detected by the host vehicle continues to move over the long-term prediction period .
The automatic driving device according to claim 1. The short-term prediction unit calculates the short-term collision risk when the short-term cost function is minimized when it is assumed that the obstacle detected by the own vehicle moves over the short-term prediction period , or the obstacle detected by the own vehicle If the short-term collision risk when the short-term cost function is minimized when the obstacle is assumed to be stationary is equal to or greater than the threshold, the very short-term prediction unit is configured to calculate the collision risk during the very short-term prediction period. instructing to predict the optimal driving behavior when the vehicle is traveling;
The automatic driving device according to claim 1 or 2. Based on the destination potential corresponding to the strength of the gravitational force toward the preset destination, which is executed by a computer, and the lane potential corresponding to the strength of the gravitational force toward the center of the lane in which the vehicle is traveling, the current Automated driving that determines a driving behavior that is a driving mode determined by a combination of a driving route and a target speed starting at a position, and a movement of the own vehicle starting from the current point determined by the acceleration/deceleration and turning angular velocity of the own vehicle. A method,
A long-term collision risk indicating the magnitude of the risk of the own vehicle colliding with an obstacle when traveling on the driving route during the long-term prediction period from the current time to the long-term prediction time, from the plurality of driving modes, and the long-term prediction period. predicting the driving mode that is optimal for the host vehicle to travel during the long-term prediction period so that the long-term cost function expressed by the distance that can be traveled is minimized;
While the own vehicle is traveling based on the predicted driving mode in the long-term prediction period, the own vehicle collides with an obstacle during a short-term prediction period from the current time to a short-term prediction time shorter than the long-term prediction time. When the host vehicle travels during the short-term prediction period, the vehicle is configured to predicting a short-term driving behavior that is the optimal driving behavior;
If the short-term collision risk when the short-term cost function is minimized is greater than or equal to a threshold, the own vehicle collides with an obstacle during a very short-term prediction period from the present to a very short-term prediction time shorter than the short-term prediction time. The driving behavior that is different from the short-term driving behavior is set during the very short-term prediction period so that the very short-term collision risk indicating the size of the risk of the collision and the very short-term cost function expressed by the lane potential are minimized. predicting the optimal driving behavior when the vehicle is traveling;
An automated driving method with The computer calculates the current position of the vehicle based on the destination potential, which corresponds to the strength of the gravitational force toward the preset destination, and the lane potential, which corresponds to the strength of the gravitational force toward the center of the lane in which the vehicle is traveling. An automatic driving method that determines a driving behavior that is a driving mode determined by a combination of a driving route starting from a starting point and a target speed, and a driving action that is a movement of the own vehicle starting from a current point determined by the acceleration/deceleration and turning angular velocity of the own vehicle. A program for executing
to the computer;
A long-term collision risk indicating the magnitude of the risk of the own vehicle colliding with an obstacle when traveling on the driving route during the long-term prediction period from the current time to the long-term prediction time, from the plurality of driving modes, and the long-term prediction period. predicting the driving mode that is optimal for the host vehicle to travel during the long-term prediction period so that the long-term cost function expressed by the distance that can be traveled is minimized;
While the own vehicle is traveling based on the predicted driving mode in the long-term prediction period, the own vehicle collides with an obstacle during a short-term prediction period from the current time to a short-term prediction time shorter than the long-term prediction time. When the host vehicle travels during the short-term prediction period, the vehicle is configured to predicting a short-term driving behavior that is the optimal driving behavior;
If the short-term collision risk when the short-term cost function is minimized is greater than or equal to a threshold, the own vehicle collides with an obstacle during a very short-term prediction period from the present to a very short-term prediction time shorter than the short-term prediction time. The driving behavior that is different from the short-term driving behavior is set during the very short-term prediction period so that the very short-term collision risk indicating the size of the risk of the collision and the very short-term cost function expressed by the lane potential are minimized. predicting the optimal driving behavior when the vehicle is traveling;
A program to run.

Images