CN110632916A

CN110632916A - Behavior prediction device and automatic driving device

Info

Publication number: CN110632916A
Application number: CN201910265673.2A
Authority: CN
Inventors: 安井裕司; 市野佑树
Original assignee: Honda Motor Co Ltd
Current assignee: Honda Motor Co Ltd
Priority date: 2018-06-22
Filing date: 2019-04-03
Publication date: 2019-12-31
Anticipated expiration: 2039-04-03
Also published as: JP2019220054A; JP7125286B2; CN110632916B

Abstract

The invention provides a behavior prediction device and an automatic driving device capable of improving the prediction precision of the behavior of a traffic participant. The action prediction device 1 calculates a predicted speed V (k + n), a predicted X coordinate value X (k + n), and a predicted Y coordinate value Y (k + n) of the traffic participant using a model (expressions (12), (21), and (22)) including the estimated acceleration α and the estimated curvature ρ of the traffic participant, calculates a speed interference function value Pv _ j and a trajectory interference function value Pt _ j representing the degree of interference with the traffic participant of an interfering object in the vicinity of the traffic participant, and determines the estimated acceleration α and the estimated curvature ρ so that the degree of interference is reduced.

Description

Behavior prediction device and automatic driving device

Technical Field

The present invention relates to a behavior prediction device and an automatic driving device for predicting a behavior of a traffic participant.

Background

Conventionally, as a behavior prediction device, a behavior prediction device described in patent document 1 is known. This behavior prediction device is a behavior prediction device that predicts the behavior of a pedestrian as a traffic participant, and includes a camera device, a yaw rate sensor, a speed sensor, and the like mounted on a vehicle. In this behavior prediction device, a predicted value of a state vector in which the speed vx, vy and the position x, y of the pedestrian in the two-dimensional coordinate system are elements is calculated as a predicted value of the behavior of the pedestrian using a linear model expression modeled by a kalman filter.

Prior art documents

Patent document

Patent document 1: japanese laid-open patent publication No. 2009-89365

Disclosure of Invention

Problems to be solved by the invention

According to the conventional behavior prediction device described above, there is a problem that, when there is an interfering object (for example, another pedestrian, a state of a signal, and a surrounding environment) that may interfere with the behavior of a pedestrian, the behavior of the pedestrian cannot be predicted in consideration of the interfering object, and the prediction accuracy is low, based on the relationship using the linear model expression modeled by the kalman filter. For the same reason, under the condition that the pedestrian bends or walks while walking, the calculation error of the predicted value increases, and the prediction accuracy significantly decreases.

The present invention has been made to solve the above-described problems, and an object thereof is to provide a behavior prediction device and an automatic driving device capable of improving the accuracy of prediction of the behavior of a traffic participant.

Means for solving the problems

In order to achieve the above object, the present invention provides a behavior prediction device 1 for predicting the behavior of traffic participants (pedestrians M, M2 to M6, vehicles 3A, 3B), comprising: a peripheral situation recognition unit (ECU2, information detection device 4, various parameter calculation unit 9) that recognizes the peripheral situation of the traffic participant; an action prediction value calculation means (ECU2, action prediction value calculation unit 10) for calculating an action prediction value (predicted speed V (k + n), predicted X coordinate value X (k + n), predicted Y coordinate value Y (k + n)) which is a prediction value of a future action of the traffic participant, using the result of recognition of the surrounding situation by the surrounding situation recognition means (position (X, Y), speed V, inclination angle θ) and an action model (equations (12), (21), (22)) which models the action of the traffic participant, the action model including action pattern parameters (estimated acceleration α, estimated curvature ρ) which indicate the action pattern of the traffic participant; an interference degree parameter calculation means (an interference function value calculation unit 8) for calculating an interference degree parameter (a velocity interference function value Pv _ j, a trajectory interference function value Pt _ j) indicating the degree of interference between an interfering object in the vicinity of a traffic participant and the traffic participant, using the action prediction value; and an action pattern parameter determination means (ECU2, acceleration calculation unit 20, curvature calculation unit 50) for determining the action pattern parameter so that the degree of disturbance indicated by the disturbance degree parameter is reduced.

According to this action prediction device, an action prediction value that is a prediction value of a future action of the traffic participant is calculated using the recognition result of the surrounding situation by the surrounding situation recognition means and an action model that models the action of the traffic participant, the action model including an action mode parameter that indicates the action mode of the traffic participant, and an interference degree parameter that indicates the degree of interference between the interference object and the traffic participant using the action prediction value. Further, since the behavior pattern parameter is determined such that the degree of interference indicated by the interference degree parameter is reduced, the predicted behavior value is calculated by using the behavior model including such behavior pattern parameter, so that the degree of interference of the traffic participant by the interfering object is reduced. As a result, the accuracy of calculating the predicted action value can be improved, and the accuracy of predicting the action of the traffic participant can be improved (the "interfering object" in the present specification includes other traffic participants, display of signals, road division, traffic environment, obstacles, and the like).

In the present invention, it is preferable that the traffic participant's action is a time-series position trajectory representing an actual spatial position of the traffic participant, the action model is a position trajectory model modeling the position trajectory of the traffic participant, and the action prediction value is a prediction value of the position trajectory, and the traffic participant control device further includes: a position trajectory acquisition unit (ECU2, various parameter calculation unit 9) that acquires the position trajectory of the traffic participant; past position trajectory storage means (ECU2, trajectory model evaluation function value calculation unit 51) for storing past position trajectories (past X 'coordinate values X' (k-i), past Y 'coordinate values Y' (k-i)) as past values of the position trajectory acquired by the position trajectory acquisition means; a past position trajectory model value calculation means (ECU2, trajectory model evaluation function value calculation unit 51) for calculating a past position trajectory model value (past trajectory model Y 'coordinate value Ym' (k-i)) as a model value of a past position trajectory using the behavior pattern parameter and the coordinate value (past X 'coordinate value X' (k-i)) as at least one component of the past position trajectory stored in the past trajectory storage means; and an error parameter calculation unit (ECU2, trajectory model evaluation function value calculation unit 51) that calculates an error parameter (sum of squares error RSS) indicating an error between the past position trajectory and the past position trajectory model value, and an action pattern parameter determination unit that determines the action pattern parameter so that the error indicated by the error parameter is further reduced in addition to the disturbance degree indicated by the disturbance degree parameter.

According to the behavior prediction device, the position trajectory of the traffic participant is acquired, the past position trajectory which is the past value of the acquired position trajectory is stored, the past position trajectory model value which is the model value of the position trajectory is calculated using the behavior pattern parameter and the coordinate value which is at least one component in the past position trajectory stored in the past trajectory storage means, and the error parameter which indicates the error between the past position trajectory and the past position trajectory model value is calculated. Furthermore, since the behavior parameter is determined such that the error indicated by the error parameter is further reduced in addition to the degree of disturbance indicated by the degree of disturbance parameter, the behavior parameter can be determined such that the correlation between the past position trajectory and the past position trajectory model value is high. As a result, by using the position trajectory model including such an action pattern parameter, the accuracy of calculating the predicted value of the position trajectory can be further improved, and the accuracy of predicting the position trajectory of the traffic participant can be further improved. In addition to this, even under the condition that the degree of interference with the traffic participants of the interfering object is extremely small without reducing the degree of interference, the action pattern parameters can be determined so that only the correlation between the past position trajectory and the past position trajectory model value becomes high (in addition, "acquisition" of "acquisition of position trajectory" or the like in the present specification is not limited to directly detecting these values by a sensor or the like, but includes calculating the values based on some parameters).

In the present invention, it is preferable that the position trajectory is a position trajectory of a two-dimensional coordinate system having the first coordinate value and the second coordinate value as components, in a past position trajectory, a first coordinate past value (past X 'coordinate value X' (k-i)) as a past value of a first coordinate value and a second coordinate past value (past Y 'coordinate value Y' (k-i)) as a past value of a second coordinate value are used as components, a past position trajectory model value calculation means calculates a second coordinate past model value (past trajectory model Y 'coordinate value Ym' (k-i)) as a model value of the second coordinate past value as a past position trajectory model value using an action mode parameter and the first coordinate past value in the past position trajectory, and an error parameter calculation means calculates a value representing an error between the second coordinate past value and the second coordinate past model value as an error parameter.

According to the behavior prediction device, the position trajectory is a position trajectory of a two-dimensional coordinate system having the first coordinate value and the second coordinate value as components, and in the past position trajectory, a first coordinate past value which is a past value of the first coordinate value and a second coordinate past value which is a past value of the second coordinate value are used as components, and a second coordinate past model value which is a past value of the second coordinate is calculated as the past position trajectory model value using the behavior pattern parameter and the first coordinate value in the past position trajectory. Further, since the error parameter is calculated as a value indicating an error between the second coordinate past value and the second coordinate past model value, when various parameters such as the error parameter are calculated, the calculation load can be reduced and the calculation time can be shortened as compared with the case of using a position trajectory model of a coordinate system having a dimension higher than two dimensions (for example, three dimensions).

In the present invention, it is preferable that the disturbance degree parameter calculation unit calculates modification values (X-coordinate-side modification coefficient Kx, Y-coordinate-side modification coefficient Kv) from the error parameter (sum of squares error RSS), and calculates the disturbance degree parameter (trajectory disturbance function value Pt _ j) using a value obtained by modifying the reference values (X-coordinate-side reference value Pt _ X _ bs, Y-coordinate-side reference value Pt _ Y _ bs) of the disturbance degree parameter with the modification values.

According to this behavior prediction device, since the modified value is calculated from the error parameter and the disturbance degree parameter is calculated using the value obtained by modifying the reference value of the disturbance degree parameter with the modified value, in the case where the error between the past position trajectory and the past position trajectory model value is large and the degree to which the past position trajectory of the traffic participant is separated from the position trajectory model is large, the disturbance degree parameter can be calculated while reflecting this, and the calculation accuracy can be improved.

In the present invention, it is preferable that the position trajectory model is a position trajectory model obtained by modeling the position trajectory of the traffic participant as an arc-shaped position trajectory in which a straight line extending in the traveling direction at the current point of the traffic participant is defined as a tangent line.

According to this behavior prediction device, since the position trajectory model is a position trajectory model obtained by modeling the position trajectory of the traffic participant as an arc-shaped position trajectory in which a straight line extending in the traveling direction at the current point of the traffic participant is taken as a tangent line, the predicted value of the position trajectory can be calculated with high accuracy based on the straight line even under the condition that the traffic participant moves while moving while curving or meandering, unlike the case of patent document 1.

In the present invention, preferably, the action mode parameter determining means determines the action mode parameter (estimated curvature ρ) as a solution when the evaluation function including the disturbance degree parameter (trajectory disturbance function value Pt _ J) and the error parameter (sum of squares error RSS) as arguments shows an extreme value, using an evaluation function (trajectory model evaluation function value J _ trj).

According to this behavior prediction device, since the evaluation function including the disturbance degree parameter and the error parameter as arguments specifies the behavior pattern parameter as a solution when the extreme value is expressed as the evaluation function, the behavior pattern parameter can be specified as an optimal solution capable of reducing the disturbance degree expressed by the disturbance degree parameter and the error expressed by the error parameter.

In the present invention, preferably, the action parameter determining means determines the action parameter as a solution when the evaluation function includes the disturbance degree parameter as an argument, using the evaluation function.

According to this behavior prediction device, since the evaluation function including the disturbance degree parameter as an argument is used to specify the behavior mode parameter as a solution when the evaluation function indicates the extremum, the behavior mode parameter can be specified as an optimal solution capable of reducing the disturbance degree indicated by the disturbance degree parameter and the error indicated by the error parameter.

In the present invention, preferably, the action of the traffic participant is at least one of a position trajectory representing a time series of actual spatial positions of the traffic participant and a speed of the traffic participant.

According to this behavior prediction device, it is possible to calculate predicted values of a time-series position trajectory representing the actual spatial position of a traffic participant and/or the speed of the traffic participant, so that the degree of interference of the traffic participant with an interfering object is reduced, and to calculate these predicted values with high accuracy.

The automatic driving device 1 of the present invention is characterized by comprising: the behavior prediction device 1 described in any one of the above; and control means (ECU2, STEP 30-32) for executing automatic drive control of the vehicle 3 based on the action prediction value (predicted speed V (k + n), predicted X-coordinate value X (k + n), and predicted Y-coordinate value Y (k + n)).

According to this automatic driving apparatus, since the predicted value of the action of the traffic participant can be calculated with high accuracy, automatic driving of the vehicle can be controlled with high accuracy based on the predicted value of the action.

Drawings

Fig. 1 is a diagram showing a configuration of an autonomous vehicle including a behavior prediction device and an autonomous device according to an embodiment of the present invention.

Fig. 2 is a block diagram showing the functional structure of the automatic driving apparatus.

Fig. 3 is a block diagram showing the configuration of the action prediction value calculation unit.

Fig. 4 is a block diagram showing the configuration of the acceleration calculation unit.

Fig. 5 is a diagram for explaining the relationship between the velocity model evaluation function value J _ v and the signal added acceleration α w.

Fig. 6 is a diagram for explaining the relationship between the moving average Pa _ v and the signal added acceleration α w.

Fig. 7 is a block diagram showing the configuration of the curvature calculating section.

Fig. 8 is a diagram for explaining a speed model of a traffic participant.

Fig. 9 is a diagram showing a trajectory before a traffic participant reaches an interfering object.

Fig. 10 is a diagram for explaining a trajectory model of a traffic participant.

Fig. 11 is a diagram showing a trajectory of a traffic participant before the traffic participant reaches an interfering object in an X '-Y' coordinate system.

Fig. 12 is a diagram showing an example of calculation of the predicted speed V (k + n), the predicted X coordinate value X (k + n), and the predicted Y coordinate value Y (k + n) when the pedestrian walks in a substantially straight line at a constant speed under the condition where no interfering object is present.

Fig. 13 is a diagram showing an example of calculation of the predicted speed V (k + n), the predicted X coordinate value X (k + n), and the predicted Y coordinate value Y (k + n) when the pedestrian walks on the homokinetic curve under the condition that no interfering object is present.

Fig. 14 is a diagram showing an example of calculation of the predicted speed V (k + n), the predicted X coordinate value X (k + n), and the predicted Y coordinate value Y (k + n) in the case where the pedestrian decelerates and walks substantially straight under the condition where no interfering object is present.

Fig. 15 is a diagram showing an example of calculation of the predicted speed V (k + n), the predicted X coordinate value X (k + n), and the predicted Y coordinate value Y (k + n) in the case where the pedestrian is accelerated and the curve walks under the condition that there is no interfering object.

Fig. 16A is a diagram showing an example of calculation of the predicted Y coordinate value Y (k + n) and the predicted X coordinate value X (k + n) under traffic conditions where traffic lights are blue signals in a state where a pedestrian walks from a sidewalk to a lane.

Fig. 16B is a diagram showing an example of calculation of the predicted speed V (k + n) under the traffic condition of fig. 16A.

Fig. 16C is a diagram showing an example of a map for calculating the speed disturbance function value Pv _1 under the traffic condition of fig. 16A.

Fig. 17A is a diagram showing an example of calculation of the predicted Y coordinate value Y (k + n) and the predicted X coordinate value X (k + n) under traffic conditions in which traffic lights are red signals in a state in which a pedestrian walks from a sidewalk toward a lane.

Fig. 17B is a diagram showing an example of calculation of the predicted speed V (k + n) under the traffic condition of fig. 17A.

Fig. 17C is a diagram showing an example of a map for calculating the speed disturbance function value Pv _1 under the traffic condition of fig. 17A.

Fig. 18A is a diagram showing an example of calculation of the predicted Y coordinate value Y (k + n) and the predicted X coordinate value X (k + n) under traffic conditions in which a vehicle is parked in the traveling direction thereof as an interfering object in a state in which a pedestrian walks on a constant velocity curve.

Fig. 18B is a diagram showing an example of calculation of the predicted speed V (k + n) under the traffic condition of fig. 18A.

Fig. 18C is a diagram showing an example of a map used for calculation of the X-coordinate map value Pt _2X and the Y-coordinate map value Pt _2Y of the trajectory interference function value Pt _2 under the traffic condition of fig. 18A.

Fig. 19A is a diagram showing an example of calculation of the predicted Y coordinate value Y (k + n) and the predicted X coordinate value X (k + n) under traffic conditions in which a lane exists as an interfering object in the traveling direction of a pedestrian walking on a constant velocity curve.

Fig. 19B is a diagram showing an example of calculation of the predicted speed V (k + n) under the traffic condition of fig. 19A.

Fig. 19C is a diagram showing an example of a map used for calculation of the trajectory interference function value Pt _3 under the traffic condition in fig. 19A.

Fig. 20A is a diagram showing an example of calculation of the predicted Y-coordinate value Y (k + n) and the predicted X-coordinate value X (k + n) under traffic conditions in which three pedestrians intersect each other.

Fig. 20B is a diagram showing an example of calculation of the predicted speed V (k + n) under the traffic condition of fig. 20A.

Fig. 21 is a diagram showing an example of a map used in the calculation of the X-coordinate side reference value Pt _ X _ bs.

Fig. 22 is a diagram for explaining when the map of the X-coordinate side reference value Pt _ X _ bs is used for actual calculation.

Fig. 23 is a diagram showing an example of a map used in the calculation of the Y-coordinate side reference value Pt _ Y _ bs.

Fig. 24 is a diagram for explaining when the map of the Y-coordinate-side reference value Pt _ Y _ bs is used for actual calculation.

Fig. 25 is a diagram showing an example of a map used for calculation of the X-coordinate-side modification coefficient Kx.

Fig. 26 is a diagram showing a state in which the mapping value of the X-coordinate-side reference value Pt _ X _ bs is modified by the X-coordinate-side modification coefficient Kx.

Fig. 27 is a diagram showing an example of a map used in the calculation of the Y-coordinate-side modification coefficient Ky.

Fig. 28 is a diagram showing a state in which the mapping value of the Y-coordinate-side reference value Pt _ Y _ bs is modified by the Y-coordinate-side modification coefficient Ky.

Fig. 29 is a diagram showing the input/output state of data between the interference function value calculation unit and the action prediction value calculation unit.

Fig. 30 is a diagram showing a simulation result of the calculation of the predicted position by the action prediction value calculation unit.

Fig. 31 is a flowchart showing the interference function value calculation processing.

Fig. 32 is a flowchart showing the automatic driving preparation process.

Fig. 33 is a flowchart showing the action prediction value calculation processing.

Fig. 34 is a diagram for explaining the travel track determination processing.

Fig. 35 is a flowchart showing the automatic driving control process.

Description of the reference numerals

1 automatic driving device and behavior prediction device

ECU (peripheral condition recognition means, predicted action value calculation means, action mode parameter determination means, position trajectory acquisition means, past position trajectory storage means, past position trajectory model value calculation means, error parameter calculation means, control means)

3 vehicle

3A vehicle (traffic participant)

3B vehicle (traffic participant)

4 information detecting device (surrounding situation recognizing means)

9 various parameter calculating part (surrounding situation recognizing means, position track acquiring means)

8 interference function value calculating part (interference degree parameter calculating unit)

10 action prediction value calculating part (action prediction value calculating means)

20 acceleration calculating part (action mode parameter determining unit)

50 curvature calculating part (action mode parameter determining means)

51 trajectory model evaluation function value calculation unit (past position trajectory storage means, past position trajectory model value calculation means, error parameter calculation means)

M, M2-M6 pedestrian (traffic participant)

X X coordinate value (recognition result of peripheral condition)

Y Y coordinate value (recognition result of peripheral condition)

V speed of traffic participants (recognition of surrounding situation)

Theta Angle of inclination (recognition of surrounding situation)

Alpha estimated acceleration (action mode parameter)

Rho estimated curvature (action mode parameter)

V (k + n) prediction speed (action prediction value)

X (k + n) prediction X coordinate value (action prediction value)

Y (k + n) prediction Y coordinate value (action prediction value)

Pv _ j speed disturbance function value (disturbance degree parameter)

Pt _ j track interference function value (interference degree parameter)

Reference value on Pt _ X _ bs X coordinate side (reference value for interference degree parameter)

Reference value on Pt _ Y _ bs Y-coordinate side (reference value for interference degree parameter)

Kx X coordinate side modification coefficient (modified value)

Ky Y coordinate side modification coefficient (modified value)

X '(k-i) past X' coordinate value (past position track, past value of first coordinate)

Y '(k-i) past Y' coordinate value (past position track, past second coordinate value)

Ym '(k-i) past trajectory model Y' coordinate value (past position trajectory model value, second coordinate past model value)

RSS sum of squares error (error parameter)

The J _ trj trajectory model evaluates the function values.

Detailed Description

Hereinafter, a behavior prediction device and an automatic driving device according to an embodiment of the present invention will be described with reference to the drawings. Since the automatic driving device of the present embodiment also serves as a behavior prediction device, the automatic driving device will be described below, and the function and structure of the behavior prediction device will be described below.

As shown in fig. 1, the automatic driving apparatus 1 is an automatic driving apparatus applied to a four-wheeled vehicle 3 (hereinafter referred to as "own vehicle 3"), and predicts the behavior of a traffic participant by an algorithm described later in order to execute automatic driving control of the own vehicle 3.

The automatic steering apparatus 1 includes an ECU2, and the information detection device 4, the motor 5, and the actuator 6 are electrically connected to the ECU 2. The information detection device 4 (surrounding situation recognition means) is configured by a camera, a millimeter wave radar, a LIDAR, a laser radar, a sonar, a GPS, various sensors, an information receiving device from an infrastructure called ITS or VTS, and the like, and detects the position of the host vehicle 3 and surrounding information of the traveling direction of the host vehicle 3 and outputs the information to the ECU 2. In this case, the surrounding information includes a signal, a boundary between a lane and a sidewalk, information of a traffic participant (a pedestrian, another vehicle, an obstacle), and the like.

As will be described later, the ECU2 predicts the behavior of the traffic participants based on the surrounding information from the information detection device 4, and determines the future travel locus, the travel speed, and the like of the host vehicle 3 based on the prediction result of the behavior of the traffic participants, the position of the host vehicle 3, the traffic environment around the host vehicle 3, and the like.

The motor 5 is constituted by, for example, an electric motor, and when the future travel locus, the travel speed, and the like of the host vehicle 3 are determined, the ECU2 controls the output of the motor 5 so that the host vehicle 3 travels along the travel locus and the travel speed.

The actuator 6 is constituted by a brake actuator, a steering actuator, and the like, and when the future travel locus, the travel speed, and the like of the host vehicle 3 are determined, the ECU2 controls the operation of the actuator 6 so that the host vehicle 3 travels along the travel locus and the travel speed.

On the other hand, the ECU2 is constituted by a microcomputer including a CPU, a RAM, a ROM, an I/O interface, various electric circuits (none of which is shown), and the like. The ECU2 executes action prediction value calculation processing and the like as described later on the basis of the peripheral information and the like from the information detection device 4. In the following description, it is assumed that various data calculated by ECU2 are stored in the RAM of ECU 2.

In the present embodiment, the ECU2 corresponds to the surrounding situation recognition means, the action prediction value calculation means, the action mode parameter specification means, the position trajectory acquisition means, the past position trajectory storage means, the past position trajectory model value calculation means, the error parameter calculation means, and the control means.

Next, a functional configuration of the automatic driving device 1 according to the present embodiment will be described with reference to fig. 2. In order to execute the automatic driving control of the host vehicle 3, the automatic driving apparatus 1 predicts the behavior of the traffic participant by the algorithm described below, and in the following description, first, a case where the participant is a pedestrian and the behavior thereof is predicted as a speed and a position trajectory will be described as an example.

In the following description, an object that interferes with the movement of a traffic participant is referred to as an "interfering object". In this case, other traffic participants and traffic environments (road environment, state of signal, and the like) correspond to the interfering object.

As shown in fig. 2, the automatic driving device 1 includes an interference function value calculation unit 8, various parameter calculation units 9, and an action prediction value calculation unit 10. The interference function value calculation unit 8 is provided as an arithmetic storage device independent of the ECU2, and the various parameter calculation units 9 and the action prediction value calculation unit 10 are specifically constituted by an ECU 2. The arithmetic processing by these elements 8 to 10 is executed at a predetermined control cycle Δ T (for example, 5 msec).

In the following description, symbol (k) in each discrete data represents a control time, symbol (k) represents a current value calculated or sampled at a current control timing, and symbol (k-1) represents a previous value calculated or sampled at a previous control timing. In the following discrete data, these symbols (k) are omitted as appropriate. In the following description, for convenience, the front-rear direction of the traffic participant is defined as an X coordinate axis, the left-right direction is defined as a Y coordinate axis, and the current position of the traffic participant is defined as the position (X (k), Y (k)).

Next, the interference function value calculation unit 8 (interference degree parameter calculation means) will be described. The interference function value calculation unit 8 calculates a velocity interference function value Pv _ j and a trajectory interference function value Pt _ j, which will be described later, based on the peripheral information, the calculation results of the various parameters by the various parameter calculation unit 9, and the calculation results of the action prediction value by the action prediction value calculation unit 10 at the control cycle Δ T, and outputs the interference function values Pv _ j and Pt _ j to the action prediction value calculation unit 10. A method of calculating the interference function values Pv _ j and Pt _ j in the interference function value calculation unit 8 will be described later.

The various parameter calculation unit 9 calculates various parameters such as the speed V of the traffic participant, the position (X, Y) of the traffic participant, and the inclination angle θ of the traveling direction using a predetermined calculation algorithm (for example, a reinforcement learning method using a deep neural network) based on the peripheral information from the information detection device 4 at the control cycle Δ T. In the present embodiment, the various parameter calculation unit 9 corresponds to the surrounding situation recognition means and the position trajectory acquisition means, and the speed V of the traffic participant, the position (X, Y) of the traffic participant, and the inclination angle θ of the traveling direction correspond to the recognition result of the surrounding situation.

Next, the action prediction value calculation unit 10 (action prediction value calculation means) will be explained. The action prediction value calculation unit 10 is an action prediction value calculation unit that calculates, as an action prediction value, a predicted speed V (k + n), a predicted X coordinate value X (k + n), a predicted Y coordinate value Y (k + n), and the like, using an action model (a speed model, a position trajectory model) of a traffic participant, which will be described later, based on the various parameters and the surrounding information. The predicted speed V (k + n) is a predicted value of the speed of the traffic participant at a future time that is n (n is an integer) control cycles apart from the current control timing, and is calculated at the current control timing.

Similarly, the predicted X-coordinate value X (k + n) and the predicted Y-coordinate value Y (k + n) are predicted values of X-coordinate values and Y-coordinate values of future traffic participants separated by n (n is an integer) control cycles from the current control timing, and are calculated at the current control timing.

These predicted X-coordinate value X (k + n) and predicted Y-coordinate value Y (k + n) correspond to predicted values of the positions of the traffic participants, and therefore, these are hereinafter referred to as predicted positions (X (k + n), Y (k + n)) as appropriate. In the present embodiment, the predicted speed V (k + n), the predicted X-coordinate value X (k + n), and the predicted Y-coordinate value Y (k + n) correspond to the action prediction value.

As shown in fig. 3, the action predicted value calculation unit 10 includes an acceleration calculation unit 20, a first predicted value calculation unit 40, a curvature calculation unit 50, and a second predicted value calculation unit 70. As will be described later, the acceleration calculation unit 20 calculates the estimated acceleration α of the traffic participant using the peripheral information, various parameters, and the like.

In addition, as will be described later, the first predicted value calculation unit 40 calculates the predicted speed V (k + n) and the predicted along-the-road distance Z (k + n) using the estimated acceleration α of the traffic participant and the like. Further, the curvature calculation unit 50 calculates the estimated curvature ρ using the predicted along-the-road distance Z (k + n) and the like, as will be described later.

Then, as will be described later, the second predicted value calculation unit 70 calculates the predicted X-coordinate value X (k + n) and the predicted Y-coordinate value Y (k + n) using the estimated curvature ρ and the like. These predicted X-coordinate value X (k + n) and predicted Y-coordinate value Y (k + n) correspond to predicted values of future positions of traffic participants, and therefore these are appropriately summarized as predicted positions (X (k + n), Y (k + n)) in the following description.

Next, the acceleration calculation unit 20 (action mode parameter determination means) will be described. The acceleration calculation unit 20 is an acceleration calculation unit that calculates an estimated acceleration α (an action mode parameter) of a traffic participant, and includes a velocity model evaluation function value calculation unit 21 and an extremum search controller 30, as shown in fig. 4.

The speed model evaluation function value calculation unit 21 models the speed V of the traffic participant and the along-the-road distance Z as described below, thereby calculating a speed model evaluation function value J _ V. First, as shown in fig. 8, when it is assumed that the traffic participant moves along an arc-shaped trajectory, a model of the speed V (i.e., a speed model) and a model of the distance Z along the road are defined by the following expressions (1) and (2), respectively.

[ expression 1]

V(k)＝V(k-1)+α(k-1)·ΔT…(j)

[ expression 2]

As is apparent from the above expression (1), in the velocity model, the estimated acceleration α corresponds to a model parameter. In this expression (1), the following expression (3) can be obtained by replacing the left-hand velocity V with the estimated velocity V _ hat and replacing the estimated acceleration α with the signal added acceleration α w. As will be described later, the signal additional acceleration α w is calculated in the extremum seeking controller 30.

[ expression 3]

V_hat(k)＝V(k-1)+αw(k-1)·ΔT…(3)

Using the estimated velocity V _ hat of this expression (3), a velocity model evaluation function value J _ V is calculated by the following expression (4).

[ expression 4]

The first term on the right side of this expression (4) is the square error of the estimated speed V _ hat and the speed V using the value calculated by the various parameter calculating section 9 at the control timing of this time.

Further, Pv _ j (X (k + n), Y (k + n)) of the second term on the right side of expression (4) is a velocity interference function value, which is a value indicating the degree of interference (degree of influence) generated between the traffic participant and the interfering object at the position when the traffic participant moves along the circular arc-shaped trajectory and reaches the predicted position (X (k + n), Y (k + n), more specifically, the degree of interference with respect to the velocity V of the traffic participant, as shown in fig. 9.

Further, in the second term on the right side of expression (4), N is an integer corresponding to the number of interfering objects, and Wv _ j is a weighting coefficient. The weighting coefficient Wv _ j is a coefficient for weighting the velocity interference function value Pv _ j, and is set to a value between 0 and 1 according to the type of the interference object.

Next, the extremum seeking controller 30 will be explained. The extremum search controller 30 is an extremum search controller that calculates the estimated acceleration α and the signal added acceleration α w using the velocity model evaluation function value J _ v, and includes, as shown in fig. 4, a cleaning filter 31, a reference signal generator 32, a multiplier 33, a moving average filter 34, a search controller 35, and a signal added acceleration calculation unit 36.

In the cleaning filter 31, a filter value H _ v is calculated by the following expression (5).

[ expression 5]

H_v(k)＝J_v(k)-J_v(k-1)…(5)

As shown in the above expression (5), the filter value H _ v is calculated as a difference value between the present time value J _ v (k) and the last time value J _ v (k-1) of the velocity model evaluation function value. The cleaning filter 31 is a cleaning filter for passing through a frequency component included in the velocity model evaluation function value J _ v and caused by a reference signal value w _ v described later. In this case, instead of expression (5), the filter value H may be calculated using a filter algorithm (butterworth high-pass filter algorithm or band-pass filter algorithm) that passes frequency components of the reference signal value w _ v, which will be described later.

Further, in the reference signal generator 32, the reference signal value w _ v is calculated by the following expression (6).

[ expression 6]

w_v(k)＝A_v·Fsin(k)…(6)

A _ v of the above expression (6) is a given gain, and Fsin is a sine function value of a given period Δ Tw _ v. As the waveform of the reference signal value, for example, a sine wave, a cosine wave, a triangular wave, a trapezoidal wave, a rectangular wave, or the like may be used.

Further, in the multiplier 33, the intermediate value Pc _ v is calculated by the following expression (7).

[ expression 7]

Pc_v(k)＝H_v(k)·w_v(k-1)…(7)

Further, in the moving average filter 34, a moving average Pa _ v is calculated by the following expression (8).

[ expression 8]

In this expression (8), in order to remove the frequency component of the reference signal value w _ v from the moving average value Pa _ v, the number of samples 1+ m _ v of the moving average value Pa _ v is set so that the product Δ T · 1+ m _ v of the number of samples 1+ m _ v and the control period Δ T becomes an integral multiple of the given period Δ Tw _ v of the sine function value Fsin.

Next, the search controller 35 calculates the estimated acceleration α by a sliding pattern control algorithm expressed by the following expressions (9) and (10).

[ expression 9]

σ_v(k)＝Pa_v(k)+S_v·Pa_v(k-1)…(9)

[ expression 10]

α(k)＝α(k-1)+K_v·σ_v(k)…(10)

S _ v of the above expression (9) is a given response specifying parameter, and σ _ v is a switching function. Further, K _ v of the above expression (10) is a given gain. As is apparent from the above expressions (9) and (10), the estimated acceleration α is calculated by the sliding pattern control algorithm that is input only by the adaptive law so as to have a function of converging the moving average Pa _ v to a value of 0.

Then, the signal added acceleration calculation unit 36 calculates the signal added acceleration α _ w by the following expression (11).

[ expression 11]

αw(k)＝α(k)+w_v(k)…(11)

Next, the reason and the principle for calculating the signal added acceleration α _ w and the estimated acceleration α using the above calculation algorithm will be described. First, as shown in the above expression (4), the velocity model evaluation function value J _ V is calculated as the sum of the right first term and the right second term by taking the square error of the estimated velocity V _ hat and the velocity V as the right first term, the sum of the multiplication values of the weighting coefficient Wv _ J and the velocity disturbance function value Pv _ J as the right second term, and the velocity model evaluation function value J _ V is calculated as the sum of the right first term and the right second term, so that in the case where the estimated acceleration α is calculated so that the velocity model evaluation function value J _ V becomes an extreme value (an extremely small value or an extremely large value), the estimated acceleration α is calculated so that the right first term and the right second term become minimum.

That is, the estimated acceleration α is calculated so that the error (deviation) between the estimated speed V _ hat and the speed V becomes minimum, and the degree of interference between the traffic participant and the interfering object also becomes minimum. For this reason, in the present embodiment, the following principle is used to calculate the estimated acceleration α so that the velocity model evaluation function value J _ v becomes an extreme value. First, based on the relationship of calculating the velocity model evaluation function value J _ v using the signal added acceleration α w, the velocity model evaluation function value J _ v represents a vibratory behavior of a given amplitude due to the characteristic (periodic function) of the reference signal value w _ v included in the signal added acceleration α w.

Here, when it is assumed that the relationship between the signal added acceleration α w and the velocity model evaluation function value J _ v is represented by a curve shown in fig. 5, for example, the oscillatory behavior of the velocity model evaluation function value J _ v caused by the reference signal value w _ v is in a state having a certain inclination as shown by arrows Y1 and Y2 in the figure. In addition, α w1 in fig. 5 is a given value of the signal added acceleration. On the other hand, the moving average Pa _ v is a moving average of the product of the filter value H _ v of the velocity model evaluation function value J _ v and the reference signal value w _ v, and therefore is a value corresponding to a correlation function between the velocity model evaluation function value J _ v and the reference signal value w _ v.

Therefore, if the moving average value Pa _ v equivalent to the correlation function is a positive value, the inclination of the velocity model evaluation function value J _ v indicates a positive value, and if the moving average value Pa _ v is a negative value, the inclination of the velocity model evaluation function value J _ v indicates a negative value. In addition to this, as for the moving average Pa _ v, calculation is performed in a state in which the frequency component of the reference signal value w _ v is removed by performing calculation by the expression (8). For the above reasons, the relationship between the moving average Pa _ v and the signal added acceleration α w can be expressed as a monotonically increasing function shown in fig. 6, for example. That is, the moving average Pa _ v indicates the direction in which the velocity model evaluation function value J _ v changes when the signal addition acceleration α w is changed.

Therefore, in order to calculate the signal added acceleration α w so that the velocity model evaluation function value J _ v becomes an extreme value (in fig. 5, a minimum value), the moving average Pa _ v may be calculated so that the inclination of the function shown in fig. 6 becomes a value 0. That is, the signal added acceleration α w, in other words, the estimated acceleration α may be calculated using a feedback control algorithm such as a sliding pattern control algorithm so that the moving average Pa _ v converges on the value 0.

Based on the above principle, in the signal addition extremum search controller 30 of the present embodiment, the estimated acceleration α is calculated as a solution in which the velocity model evaluation function value J _ v becomes an extremum using the sliding mode control algorithm of expressions (9) and (10).

Next, the first predicted value calculating unit 40 calculates the predicted speed V (k + n) and the predicted along-the-road distance Z (k + n) by model expressions shown in the following expressions (12) and (13), respectively.

[ expression 12]

V(k+n)＝V(k-1)+α(k-1)·ΔT·n…(12)

[ expression 13]

In addition, the above expression (12) is derived from the above model expression (1) of the speed V, and the above expression (13) is derived from the above model expression (2) of the along-road distance Z.

Next, the curvature calculating unit 50 (action mode parameter specifying means) will be described. The curvature calculation unit 50 is a curvature calculation unit that calculates an estimated curvature ρ (an action mode parameter) of the position trajectory of the traffic participant, and includes a trajectory model evaluation function value calculation unit 51 and an extremum search controller 60, as shown in fig. 7. In the present embodiment, the trajectory model evaluation function value calculation unit 51 corresponds to a past position trajectory storage means, a past position trajectory model value calculation means, and an error parameter calculation means.

The trajectory model evaluation function value calculation unit 51 calculates the trajectory model evaluation function value J _ trj by modeling the position trajectory of the traffic participant as a position trajectory model as described below. First, as shown in fig. 10, an arc having a straight line extending in the traveling direction of the current point of the traffic participant as a tangent is defined, and when the trajectory on the arc is assumed to be a trajectory in which the traffic participant moves from the position of the current point along the road distance Z (k) to the position of the predicted along-the-road distance Z (k + n), the movement angle is defined

Is defined by the following expression (14).

[ expression 14]

R in the expression (14) is a curvature radius, and is calculated as the reciprocal 1/ρ of the estimated curvature ρ. The reason why the previous value r (k-1) is used as the curvature radius r in this expression (14) is that the present value of the estimated curvature ρ is calculated after the calculation of the trajectory model evaluation function value J _ trj.

Further, in the case where an X '-Y' coordinate system is defined as a coordinate system having the traveling direction of the current point of the traffic participant as an X 'coordinate and the orthogonal coordinate thereof as a Y' coordinate, model expressions of the predicted X 'coordinate value X' (k + n) and the predicted Y 'coordinate value Y' (k + n) in the X '-Y' coordinate system are defined in accordance with the following expressions (15) and (16).

[ expression 15]

X’(k+n)＝r(k-1)·sin(φ(k))…(15)

[ expression 16]

Y’(k+n)＝r(k-1)-r(k-1)·cos(φ(k))…(16)

Here, when the current position of the traffic participant is used as the origin and the rotational coordinate system obtained by rotating the X ' -Y ' coordinate system by the inclination angle θ (k) in the traveling direction of the traffic participant counterclockwise in fig. 10 is used as the X ″ -Y ' coordinate system, the predicted X ' coordinate value X ' (k + n) and the predicted Y ' coordinate value Y ' (k + n) are used, and the predicted X "coordinate value X" (k + n) and the predicted Y "coordinate value Y" (k + n) are defined according to the following expressions (17) and (18). In addition, the various parameter calculation units 9 calculate the inclination angle θ (k) of the traveling direction of the traffic participant.

[ expression 17]

X”(k+n)＝X’(k+n)·cosθ(k)-Y’(k+n)·sinθ(k)…(17)

[ expression 18]

Y”(k+n)＝X’(k+n)·sinθ(k)+Y’(k+n)·cosθ(k)…(18)

Further, using the predicted X "coordinate value X" (k + n) and the predicted Y "coordinate value Y" (k + n), the predicted X coordinate value X (k + n) and the predicted Y coordinate value Y (k + n) are defined according to the following expressions (19) and (20).

[ expression 19]

X(k+n)＝X”(k+n)-X(k)…(19)

[ expression 20]

Y(k+n)＝Y’(k+n)-Y(k)…(20)

From the above expressions (19), (20) and the above expressions (17), (18), the following expressions (21), (22) are obtained as model expressions of the predicted X-coordinate value X (k + n) and the predicted Y-coordinate value Y (k + n).

[ expression 21]

X(k+n)＝X’(k+n)·cosθ(k)-X(k)-Y’(k+n)·sinθ(k)…(21)

[ expression 22]

Y(k+n)＝X’(k+n)·sinθ(k)+Y’(k+n)·cosθ(k)-Y(k)…(22)

Since the model expressions (21) and (22) are derived using the above expressions (14) to (20), the model expressions are models having the estimated curvature ρ as a model parameter.

In addition, in the case where the traffic participant is traveling straight, the estimated curvature ρ may be calculated as a value of 0, and in such a case, the predicted X coordinate value X (k + n) and the predicted Y coordinate value Y (k + n) are calculated by another calculation method instead of the expressions (21) and (22). For example, the predicted X-coordinate value X (k + n) and the predicted Y-coordinate value Y (k + n) are calculated based on the current value X (k) of the speed V, X coordinate values of the traffic participant and the current value Y (k) of the Y-coordinate value.

Further, based on the above model expressions (15), (16), (21), and (22), a calculation expression of the trajectory model evaluation function value J _ trj is derived in accordance with the following expression (23).

[ expression 23]

In the above expression (23), m of the first term on the right is an integer representing the number of samples, Ym ' (ki) is a past value of the trajectory model Y ' coordinate corresponding to the past X ' coordinate value (hereinafter referred to as "past trajectory model Y ' coordinate value"), and Y ' (k-i) is a past value of the Y ' coordinate in the past position trajectory of the actual traffic participant (hereinafter referred to as "past Y ' coordinate value"). Specific calculation methods of these values Ym '(k-i) and Y' (k-i) will be described later.

In addition, since the first term on the right side of the expression (23) corresponds to the sum of squares error, i.e., the sum of squares of the past trajectory model Y 'coordinate value Ym' (k-i) and the past Y 'coordinate value Y' (k-i), i.e., the residual sum of squares error RSS (error parameter).

Further, as will be described later, the trajectory interference function value Pt _ j (X (k + n), Y (k + n)) of the second term on the right side of expression (23) is calculated in the interference function value calculation section 8. The trajectory interference function value Pt _ jj (X (k + n), Y (k + n)) is a trajectory interference function representing the degree of interference (degree of influence) generated between the traffic participant and the interfering object at the predicted position (X (k + n), Y (k + n)) when the traffic participant arrives at the position, more specifically, the degree of interference with respect to the trajectory of the traffic participant. In the present embodiment, the trajectory interference function value Pt _ j corresponds to an interference level parameter.

Further, Wtrj _ j of the second term on the right side of the expression (23) is a weighting coefficient for weighting the interference function value Pt _ j (X (k + n), Y (k + n)), and is set to a value between 0 and 1 depending on the type of the interference object.

Next, a method of calculating the past trajectory model Y 'coordinate value Ym' (k-i) and the past Y 'coordinate value Y' (k-i) will be described. First, the current position of the traffic participant is set as the origin of the X ' -Y ' coordinate system, and the X (k-1) and Y (k-1) coordinate values and the previous values of the X ' coordinate value and the Y ' coordinate value are converted into the values of the X ' -Y ' coordinate system, thereby calculating the X ' (k-1) and Y ' (k-1) coordinate values and the previous values of the Y ' coordinate value.

Then, the last value Ym '(k-1) of the coordinate value of the trajectory model Y' is calculated by searching a map, not shown, from the last value X '(k-1) of the coordinate value of X' and the last value ρ (k-1) of the estimated curvature. The last value Ym ' (k-1) of the trajectory model Y ' coordinate value corresponds to the Y ' coordinate value on the circular arc-shaped position trajectory model shown in fig. 11 corresponding to the last value X ' (k-1) of the X ' coordinate value.

According to the same method as described above, the past X 'coordinate value X' (k-i) and the past Y 'coordinate Y' (k-i) are calculated, and a map, not shown, is searched for from the past X 'coordinate value X' (k-i) and the last value ρ (k-1) of the estimated curvature, thereby calculating the past trajectory model Y 'coordinate value Ym' (ki).

In the present embodiment, the past X 'coordinate value X' (k-i) corresponds to the past position trajectory and the first coordinate past value, the past Y 'coordinate value Y' (k-i) corresponds to the past position trajectory and the second coordinate past value, and the past trajectory model Y 'coordinate value Ym' (k-i) corresponds to the past position trajectory model value and the second coordinate past model value.

Next, the extremum seeking controller 60 will be explained. The extremum search controller 60 is an extremum search controller that calculates the estimated curvature ρ and the signal added curvature ρ w using the trajectory model evaluation function value J _ trj, and as shown in fig. 7, includes a cleaning filter 61, a reference signal generator 62, a multiplier 63, a moving average filter 64, a search controller 65, and a signal added acceleration calculation unit 66.

In this cleaning filter 61, a filter value H _ t is calculated by the following expression (24).

[ expression 24]

H_t(k)＝J_trj(k)-J_trj(k-1)…(24)

As shown in the above expression (24), the filter value H _ t is calculated as a difference between the present value J _ trj (k) and the previous value J _ trj (k-1) of the trajectory model evaluation function value. The cleaning filter 61 is a cleaning filter for passing through a frequency component included in the trajectory model evaluation function value J _ trj and caused by a reference signal value w _ t described later.

Further, in the reference signal generator 62, the reference signal value w _ t is calculated by the following expression (25).

[ expression 25]

w_t(k)＝A_t·Fsin(k)…(25)

A _ t of the above expression (25) is a given gain.

Further, in the multiplier 63, the intermediate value Pc _ t is calculated by the following expression (26).

[ expression 26]

Pc_t(k)＝H_t(k)·w_t(k-1)…(26)

Further, in the moving average filter 64, a moving average Pa _ t is calculated by the following expression (27).

[ expression 27]

In this expression (27), in order to remove the frequency component of the reference signal value w _ T from the moving average value Pa _ T, the number of samples 1+ m _ T of the moving average value Pa _ T is set so that the product Δ T · 1+ m _ T of the number of samples 1+ m _ T and the control period Δ T becomes an integral multiple of the given period Δ Tw _ T of the sine function value Fsin.

Next, the search controller 65 calculates the estimated curvature ρ by a sliding pattern control algorithm expressed by the following expressions (28) and (29).

[ expression 28]

σ_t(k)＝Pa_t(k)+S_t·Pa_t(k-1)…(28)

[ expression 29]

ρ(k)＝ρ(k-1)+K_t·σ_t(k)…(29)

S _ t of the above expression (28) is a given response specifying parameter, and σ _ t is a switching function. Further, K _ t of the above expression (29) is a given gain. As is apparent from the above expressions (28) and (29), the estimated curvature ρ is calculated by a sliding pattern control algorithm that is input only by the adaptive law so as to have a function of converging the moving average Pa _ t to a value of 0.

Then, the signal added acceleration calculation unit 66 calculates the signal added curvature ρ _ w by the following expression (30).

[ expression 30]

ρ_w(k)＝ρ(k)+w_t(k)…(30)

As described above, the extremum search controller 60 calculates the signal added curvature ρ _ w and the estimated curvature ρ by the same principle calculation method as the above-described calculation method of the signal added acceleration α w and the estimated acceleration α.

Thus, the estimated curvature ρ is calculated as a solution in which the trajectory model evaluation function value J _ trj becomes an extreme value (minimum value or maximum value), so that the square sum error RSS (right first term) and right second term of the expression (21) are calculated to be minimum. That is, the estimated curvature ρ is calculated so that an error (deviation) between the trajectory model Y 'coordinate value Ym' and the Y 'coordinate value Y' becomes minimum, and the degree of interference between the traffic participant and the interfering object also becomes minimum.

Further, the second predicted value calculating unit 70 calculates the predicted X-coordinate value X (k + n) and the predicted Y-coordinate value Y (k + n) by the above-described expressions (21) and (22) using the estimated curvature ρ and the like calculated as described above.

Next, an exemplary calculation result of the predicted speed V (k + n), the predicted X coordinate value X (k + n), and the predicted Y coordinate value Y (k + n) of the traffic participant when the above-described calculation method is used will be described. First, with reference to fig. 12 to 15, a case where only one pedestrian M2 exists as a traffic participant and no interfering object exists will be described as an example.

First, when the pedestrian M2 walks at a constant speed and in a substantially straight line, the predicted speed V (k + n), the predicted X coordinate value X (k + n), and the predicted Y coordinate value Y (k + n) are calculated as shown in fig. 12. In the case of fig. 12, the relationship between the vertical axis and the horizontal axis of fig. 10 is set to the X coordinate and the Y coordinate in order to facilitate understanding.

In fig. 12, the points indicated by o indicate the calculated values at the respective control times, and the velocity V indicated by o at the control time k + n indicates the predicted velocity V (k + n). The X-coordinate value X and the Y-coordinate value Y indicated by o at the control time k + n indicate a predicted X-coordinate value X (k + n) and a predicted Y-coordinate value Y (k + n), respectively.

Further, the trajectory and the speed indicated by the thin solid line are the actual trajectory and the speed of the pedestrian M2, respectively, and the trajectory and the speed indicated by the thick solid line are the trajectory model value and the speed model value calculated based on the estimated curvature ρ and the estimated acceleration α, respectively. The above points are also the same in the following fig. 13 to 20.

Note that the predicted speed V (k + n), the predicted X-coordinate value X (k + n), and the predicted Y-coordinate value Y (k + n) are calculated as shown in fig. 13 when the pedestrian walks in a constant velocity curve, as shown in fig. 14 when the pedestrian decelerates and walks in a substantially straight line, and as shown in fig. 15 when the pedestrian accelerates and walks in a curve.

Under the condition that there is no interfering object as shown in fig. 12 to 15 above, in the above-described calculation expressions (4) and (23) of the velocity model evaluation function value J _ v and the trajectory model evaluation function value J _ trj, the calculation is performed such that the second term on the right becomes a value of 0 and the first term on the right becomes the minimum. That is, the calculation is performed so that the square error of the estimated velocity V _ hat and the velocity V becomes minimum, and the square sum error RSS of the past trajectory model Y 'coordinate value Ym' (k-i) and the past Y 'coordinate value Y' (ki) becomes minimum.

Next, a method of calculating the velocity interference function value Pv _ j and the trajectory interference function value Pt _ j when an interfering object is present in the traveling direction of the pedestrian M2 and the principle thereof will be described with reference to fig. 16A to 19C. In this case, the interference function value calculation unit 8 calculates the current values of the velocity interference function value Pv _ j and the trajectory interference function value Pt _ j using the predicted X coordinate value X (k + n) and the predicted Y coordinate value Y (k + n) calculated by the action prediction value calculation unit 10 at the previous control timing.

In the following description, an example of calculating the predicted speed V (k + n), the predicted X coordinate value X (k + n), and the predicted Y coordinate value Y (k + n) when the calculation results of the speed disturbance function value Pv _ j and the trajectory disturbance function value Pt _ j are input from the disturbance function value calculation unit 8 to the action prediction value calculation unit 10 will be described together.

First, as shown in fig. 16A, under the condition that a pedestrian M2 walks from the sidewalk toward the crosswalk of the lane and a traffic light exists as an interfering object, when the traffic light is a blue signal, the map shown in fig. 16C is searched based on the predicted X coordinate value X (k + n), thereby calculating the velocity interference function value Pv _1 when the traffic light of the blue signal is an interfering object. As shown in this figure, in this map, the traffic light of the blue signal does not interfere with the velocity V of the pedestrian M2, and therefore the velocity interference function value Pv _1 is set to a value of 0 with respect to the X coordinate value.

Thus, when the traffic light is a blue signal, the velocity interference function value P1_ V is calculated as a value 0 based on the predicted X-coordinate value X (k + n), and the velocity interference function value Pv _1 having such a value 0 is input from the interference function value calculation unit 8 to the action prediction value calculation unit 10, whereby the predicted velocity V (k + n) is calculated so that V (k + n) ═ V (k) is satisfied as shown in fig. 16B.

For the same reason as described above, the map value of the trajectory interference function value Pt _1 when the traffic light of the blue signal is the interference target is also set to the value 0 with respect to the X-coordinate value. Thus, by inputting the trajectory interference function value Pt _1 of the value 0 from the interference function value calculation section 8 to the action prediction value calculation section 10, the trajectory model evaluation function value J _ trj is calculated so that the second term on the right of the expression (23) becomes the value 0 and the square sum error RSS of the first term on the right becomes minimum. As a result, as shown in fig. 16A, the predicted Y-coordinate value Y (k + n) and the predicted X-coordinate value X (k + n) are calculated as positions moved into the crosswalk.

On the other hand, as shown in fig. 17A, when the traffic light is a red signal, since the traffic light interferes with the speed V of the pedestrian M2, as shown in the map of fig. 17C, the speed interference function value Pv _ j when the traffic light of the red signal is an interference target is set to be a larger value as the X coordinate value becomes a value close to the boundary between the sidewalk and the lane. Then, by searching such a map based on the predicted X-coordinate value X (k + n), the velocity interference function value Pv _ j is calculated.

Further, the second right term of the calculation expression (4) of the velocity model evaluation function value J _ v is calculated by inputting the thus calculated velocity interference function value Pv _ J from the interference function value calculation unit 8 to the action prediction value calculation unit 10. In this case, since the weighting coefficient wv _1 when the traffic light is the interference target is set to a relatively large value in the motion prediction value calculation unit 10, the second term on the right side of the calculation expression (4) increases.

Further, the error between the estimated velocity V _ hat (k) and the velocity V (k) becomes large, and therefore the first term on the right side of the calculation expression (4) of the velocity model evaluation function value J _ V also becomes large. As a result, as shown in fig. 17B, the predicted speed V (k + n) is calculated so as to decrease to a value of 0. That is, the pedestrian M2 is predicted to stop.

For the same reason, the trajectory interference function value Pt _1 when the traffic light of the red signal is an interference object is set to be a larger value as the X-coordinate value becomes a value close to the boundary between the sidewalk and the lane. Then, by searching such a map based on the predicted X-coordinate value X (k + n), the trajectory interference function value Pt _1 is calculated. Further, the second right term of the calculation expression (23) of the trajectory model evaluation function value J _ trj is calculated by inputting the trajectory interference function value Pt _1 thus calculated from the interference function value calculation unit 8 to the action prediction value calculation unit 10.

In such a case, when the traffic light is a disturbance target, the motion prediction value calculation unit 10 sets the weighting coefficient wt _1 to a value smaller than the weighting coefficient wv _1 because the trajectory of the pedestrian is less influenced by the traffic light (the degree of disturbance) than the speed. As a result, as shown in fig. 17A, the predicted position (X (k + n), Y (k + n)) is calculated as a position near the boundary between the pedestrian path and the lane.

Further, as shown in fig. 18A, when a parked vehicle exists as an interfering object in the traveling direction under the condition that the pedestrian M2 walks on a constant speed curve, the interference function value calculation unit 8 calculates a trajectory interference function value Pt _2 when the parked vehicle is an interfering object, as described below. Note that the curve indicated by the broken line in fig. 18A is a curve indicating a position trajectory model when it is assumed that no interfering object is present at control time k. This point is also the same in fig. 19A and 20A described below.

In this case, as shown in the map of fig. 18C, the X-coordinate map value Pt _2X of the trajectory interference function value Pt _2 shows the maximum value at the center portion of the parked vehicle with respect to the X-coordinate value, and is set to the same tendency as the probability distribution state of the probability variable. Further, the Y-coordinate mapping value Pt _2Y of the trajectory interference function value Pt _2 represents the maximum value in the center portion of the parked vehicle with respect to the Y-coordinate value, and is set to the same tendency as the probability distribution state of the probability variable.

Then, an X-coordinate map value Pt _2X is calculated by searching the map from the predicted X-coordinate value X (k + n), a Y-coordinate map value Pt _2Y is calculated by searching the map from the predicted Y-coordinate value Y (k + n), and a track disturbance function value Pt _2 is calculated by multiplying the two values (Pt _2 ═ Pt _2X · Pt _ 2Y).

When the thus calculated trajectory interference function value Pt _2 is input from the interference function value calculation unit 8 to the action prediction value calculation unit 10, the action prediction value calculation unit 10 calculates the estimated curvature ρ such that the circular arc trajectory of the pedestrian is inverted via a straight line extending in the traveling direction of the pedestrian. As a result, as shown in fig. 18A, the predicted position (X (k + n), Y (k + n)) is calculated to be a value that can avoid parking the vehicle.

Further, in the interference function value calculation unit 8, the velocity interference function value Pv _2 when the parked vehicle is the interfering object is calculated, similarly to the trajectory interference function value Pt _ 2. When the thus calculated speed interference function value Pv _2 is input from the interference function value calculation unit 8 to the action prediction value calculation unit 10, the action prediction value calculation unit 10 calculates the predicted speed V (k + n) so as to avoid parking the vehicle, as shown in fig. 18B.

On the other hand, as shown in fig. 19A, when the lane exists as the interfering object in the traveling direction under the condition that the pedestrian M2 walks on the constant velocity curve, the interference function value calculation unit 8 calculates the trajectory interference function value P3_ t when the lane is the interfering object, as described below.

That is, as shown in the map of fig. 19C, the map value of the trajectory interference function value P3_ t is set so that the position Yref closer to the boundary between the lane and the sidewalk rapidly increases with respect to the Y coordinate value. Then, the map is searched based on the predicted Y coordinate value Y (k + n), thereby calculating a trajectory interference function value P3 — t.

When the trajectory interference function value P3_ t thus calculated is input from the interference function value calculation unit 8 to the action prediction value calculation unit 10, the estimated curvature ρ is calculated such that the circular arc trajectory of the pedestrian is inverted via a straight line extending in the traveling direction of the pedestrian. As a result, as shown in fig. 19A, the predicted position (X (k + n), Y (k + n)) is calculated as a position within the pedestrian lane, not a position within the lane.

Although not shown, the interference function value calculation unit 8 calculates a speed interference function value P3_ v when the lane is an interference object, similarly to the trajectory interference function value P3_ t. When the thus calculated speed interference function value P3_ V is input from the interference function value calculation unit 8 to the action prediction value calculation unit 10, the action prediction value calculation unit 10 calculates the predicted speed V (k + n) so that the vehicle can be prevented from being parked, as shown in fig. 19B.

Fig. 16A to 19C show an example in which the traffic participant is a single pedestrian M2, but in an actual traffic environment, a plurality of pedestrians exist as traffic participants and interfere with each other. For example, as shown in fig. 20A, under the condition that there are three pedestrians M3 to M5, the pedestrian M3 and the pedestrian M4 become mutual interference objects, and the pedestrian M4 and the pedestrian M5 become mutual interference objects.

Hereinafter, a method of calculating the disturbance function value in a case where the trajectory disturbance function value Pt _ j is given as an example and two pedestrians are mutually the targets of disturbance will be described with reference to fig. 21 to 26. Here, assuming that the predicted position (X (k + n), Y (k + n)) of one of the two pedestrians is the predicted position (X #, Y #) and the predicted position (X (k + n), Y (k + n)) of the other is the predicted position (X #, Y #), the trajectory interference function value Pt _ j (X #, Y #) in the predicted position (X #, Y #) is calculated by the following expression (31).

[ expression 31]

Pt_j(X^#，Y^#)＝Kx·Pt_x_bs(X^#-X^*)·Ky·Pt_y_bs(Y^#-Y^*)…(31)

Pt _ X _ bs of the above expression (31) is a reference value on the X-coordinate side of the trajectory interference function value (reference value of interference degree parameter). The X-coordinate-side reference value Pt _ X _ bs is set to have the same tendency as the probability function representing the probability distribution as shown in the map of fig. 21 with respect to the X-coordinate value. When the trajectory interference function value Pt _ j (X #, Y #) at the predicted position (X #, Y #) is calculated, the map shown in fig. 22 is used, and the map is searched for the value X # -X #, thereby calculating the value Pt _ X _ bs (X # -X #), and in the map shown in fig. 22, the value 0 of the X coordinate in fig. 21 is replaced with the predicted X coordinate value X #.

Further, Pt _ Y _ bs of expression (31) is a reference value on the Y-coordinate side of the trajectory interference function value (reference value of interference degree parameter). The Y-coordinate-side reference value Pt _ Y _ bs is set to have the same tendency as the probability function representing the probability distribution as shown in the map of fig. 23 with respect to the Y-coordinate value. When the trajectory interference function value Pt _ j (X #, Y #) at the predicted position (X #, Y #) is calculated, the map shown in fig. 24 is used, and the map is searched for from the value Y # -Y #, thereby calculating the value Pt _ Y _ bs (Y # -Y #), and in the map shown in fig. 24, the value 0 of the Y coordinate in fig. 23 is replaced with the predicted Y coordinate value Y #.

Further, Kx of expression (31) is an X-coordinate-side modification coefficient (modified value) which is calculated by retrieving the map shown in fig. 25 from the sum of squares error RSS. As shown in the figure, the larger the sum-of-squares error RSS, the larger the value of the X-coordinate-side modification coefficient Kx is set. The X-coordinate-side modification coefficient Kx is set in this way for the following reason.

That is, as shown in expression (31), the X-coordinate-side modification coefficient Kx is multiplied by the X-coordinate-side reference value Pt _ X _ bs, and thus as shown in fig. 26, the map value of fig. 21 or 22 (the value indicated by the broken line in fig. 26) is modified to the value indicated by the solid line in the figure. As described above, the larger the sum-of-squares error RSS, the larger the value of the X-coordinate-side modification coefficient Kx is calculated, and therefore in the case where the sum-of-squares error RSS is large, the modification is performed so that the region of the probability distribution is enlarged as compared with the case where the sum-of-squares error RSS is small.

Here, the state where the square sum error RSS is large is based on the relationship where the degree of separation between the past trajectory model Y 'coordinate value Ym' (k-i) and the past Y 'coordinate value Y' (k-i) is large, and in such a state, the possibility that the degree of interference between the traffic participant and another traffic participant increases. In this case, the state of a large degree of separation means, for example, when the pedestrian looks at the smartphone and walks, or the pedestrian walks in a drunk state, or walks in a state of holding an umbrella to look down without looking at the front. Therefore, in order to reflect this on the track disturbance function value Pt _ j (X #, Y #), the X-coordinate-side modification coefficient Kx is set as shown in fig. 25 with respect to the sum-of-squares error RSS.

Further, Ky of the equation (31) is a Y-coordinate-side modification coefficient (modification value), which is calculated by retrieving the map shown in fig. 27 from the aforementioned square sum error RSS. As shown in the figure, the larger the square sum error RSS is, the larger the Y-coordinate-side modification coefficient is set to be, as is the X-coordinate-side modification coefficient Kx. This is for the reasons described above. Further, as shown in equation (31), this Y-coordinate-side modification coefficient Ky is multiplied by the Y-coordinate-side reference value Pt _ Y _ bs, and thus the map of fig. 23 or 24 is modified as in the map shown in fig. 28.

With the above method, the disturbance function value calculation unit 8 calculates the trajectory disturbance function value Pt _ j when two pedestrians are mutually the targets of disturbance. Further, the velocity disturbance function value Pv _ j in the case where two pedestrians are mutually the disturbance target is also calculated by the same method as the trajectory disturbance function value Pt _ j.

In addition, since there are a plurality of traffic participants and interference objects in an actual traffic environment, as shown in fig. 29, the interference function value calculation unit 8 calculates 2 × N interference function values Pt _1 to N, Pv _1 to N for each control cycle using the previous value of the predicted position (X (k + N), Y (k + N)) input from the action prediction value calculation unit 10, and outputs the calculated values to the action prediction value calculation unit 10.

On the other hand, using the 2 × N interference function values Pt _ j and Pv _ j (j is 1 to N), the action prediction value calculation unit 10 calculates Nx (Nx is an integer) prediction positions (X (k + N), Y (k + N)), and outputs the calculation result to the action prediction value calculation unit 10 as the previous value of the prediction position (X (k + N), Y (k + N)). In this case, the value Nx represents the number of traffic participants.

Next, a simulation result of calculation of the predicted position (X (k + n), Y (k + n)) by the action prediction value calculation unit 10 of the present embodiment configured as described above (hereinafter referred to as "present application result") will be described with reference to fig. 30. In the figure, a curve indicated by a solid line represents the result of the present application as a position trajectory, and a curve indicated by a one-dot chain line represents an actual position trajectory of a pedestrian. For comparison, the curve indicated by the broken line is a curve representing a result of simulation of calculation of a predicted position by the same method as patent document 1 (hereinafter referred to as "comparison result") as a position trajectory.

As shown in the figure, it is known that the prediction accuracy is low because the interfering object is avoided by shifting the actual position trajectory of the pedestrian, and when the position trajectory is compared, the interfering object is avoided by shifting the actual position trajectory so as to collide with the interfering object. In contrast, in the case of the position trajectory as a result of the present application, the disturbance target can be appropriately avoided as in the case of the actual trajectory of the pedestrian, and the prediction accuracy thereof is found to be higher than the comparison result.

Next, the interference function value calculation process will be described with reference to fig. 31. The interference function value calculation process is a process of calculating the velocity interference function value Pv _ j and the trajectory interference function value Pt _ j by the above-described calculation method, and is executed by the interference function value calculation unit 8 at the predetermined control period Δ T.

As shown in this figure, first, Nx predicted positions (X (k + n), Y (k + n)) input from the action predicted value calculation unit 10, i.e., the ECU2 are read (fig. 31/STEP 1).

Next, N velocity interference function values Pv _ j (j is 1 to N) and N trajectory interference function values Pt _ j (fig. 31/STEP2) are calculated by the above-described various map search or the above-described expression (31) according to the type of the interfering object.

Next, the velocity disturbance function value Pv _ j and the trajectory disturbance function value Pt _ j calculated as described above are output to the ECU2 (fig. 31/STEP 3). After that, the present process is ended.

Next, the automatic driving preparation calculation process will be described with reference to fig. 32. This automatic driving preparation calculation process is a process of determining the travel locus of the own vehicle 3, and is executed by the ECU2 at the given control cycle Δ T.

As shown in this figure, first, action prediction value calculation processing is executed (STEP 10/fig. 32). The action prediction value calculation process is a process of calculating the predicted speed V (k + n), the predicted X-coordinate value X (k + n), and the predicted Y-coordinate value Y (k + n), and is specifically executed as shown in fig. 33.

As shown in this figure, first, the N velocity interference function values Pv _ j and the N trajectory interference function values Pt _ j input from the interference function value calculation unit 8 are read (fig. 33/STEP 20).

Next, the velocity model evaluation function value J _ v is calculated by the above expressions (3) and (4) (fig. 33/STEP21), and thereafter, the estimated acceleration α and the signal added acceleration α w are calculated by the above expressions (5) to (11) (fig. 33/STEP 22).

Next, the predicted speed V (k + n) and the predicted along-the-road distance Z (k + n) are calculated by the above equations (12) and (13) (fig. 33/STEP23), and then, the trajectory model evaluation function value J _ trj is calculated by the above expression (23) (fig. 33/STEP 24).

Next, the estimated curvature ρ and the signal-added curvature ρ w are calculated by the above-described expressions (24) to (30) (fig. 33/STEP25), and then Nx predicted X-coordinate values X (k + n) and Nx predicted Y-coordinate values Y (k + n) are calculated by the above-described expressions (21) and (22) (fig. 33/STEP 26).

Then, the Nx predicted X-coordinate values X (k + n) and the Nx predicted Y-coordinate values Y (k + n) calculated as described above are output to the interference function value calculation unit 8 (fig. 33/STEP27), and then the present process is ended.

Returning to fig. 32, after the action prediction value calculation process (fig. 32/STEP10) is executed as described above, the travel track determination process (fig. 32/STEP11) is executed. This travel locus is used to determine a future travel locus of the own vehicle 3, and specifically, the travel locus is determined based on the present vehicle speed and position of the own vehicle 3 and the predicted speed V (k + n) and the predicted position (X (k + n), Y (k + n)) of the traffic participant calculated as described above.

For example, as shown in fig. 34, when the host vehicle 3 turns right, first, the travel locus at the time of turning right of the host vehicle 3 is determined. Specifically, for example, the travel locus is determined by a method which has been proposed by the present applicant in japanese patent application No. 2018 and 3309.

Next, it is determined whether the travel locus of the host vehicle 3 and the locus of the traffic participant intersect each other based on the predicted positions (X (k + n), Y (k + n)) of the opposing vehicles 3A, 3B and the pedestrian M6 as the traffic participants. As a result, when it is determined that the trajectories of the oncoming vehicle 3A and the host vehicle 3 intersect at the point P _ CRS, the time To until the oncoming vehicle 3A reaches the point P _ CRS is calculated based on the predicted speed V (k + n) of the oncoming vehicle 3A, and the time To is compared with the time Te until the host vehicle 3 reaches the point P _ CRS.

When Te is smaller than To, the determined travel locus is used as it is. On the other hand, in the other cases, the determination result of the travel trajectory is deleted, and it is determined that the vehicle is parked at the position. As described above, after the travel track determination processing is executed, the present processing is ended.

Next, the automatic driving control calculation process will be described with reference to fig. 35. This automated driving control calculation process is a process for executing automated driving control of the host vehicle 3, and is executed by the ECU2 at a control cycle Δ Tn (for example, several tens to several hundreds of msec) longer than the predetermined control cycle Δ T.

As shown in this figure, first, the travel locus specified by the travel locus specifying processing described above is read (fig. 35/STEP 30).

Next, the motor 5 is driven based on the travel path and the current vehicle speed (fig. 35/STEP 31). Next, the actuator 6 is driven based on the travel locus and the current vehicle speed (fig. 35/STEP 32). After that, the present process is ended. As described above, by executing the automatic driving control calculation process, the own vehicle 3 travels with the travel locus described above.

As described above, according to the automatic driving device 1 of the present embodiment, the predicted speed V (k + n) is calculated using the model expression (12) that models the speed of the traffic participant, and the predicted X-coordinate value X (k + n) and the predicted Y-coordinate value Y (k + n) are calculated using the model expressions (21) and (22) that model the position trajectory of the traffic participant. Further, a velocity interference function value Pv _ j and a trajectory interference function value Pt _ j are calculated based on the type of the interfering object, the predicted X-coordinate value X (k + n), and the predicted Y-coordinate value Y (k + n).

Further, the estimated acceleration α is calculated as a solution when the velocity model evaluation function value J _ V, which includes the sum of the squared error of the estimated velocity V _ hat and the velocity V and the product Wv _ J · Pv _ J of the weighting coefficient and the velocity interference function value, represents an extreme value, and therefore, by calculating the predicted velocity V (k + n) using the model expression (12) including such an estimated acceleration α, the predicted velocity V (k + n) can be calculated as a value at which the degree of interference between the traffic participant and the interfering object becomes minimum and the error between the estimated velocity V _ hat and the velocity V becomes minimum.

Further, the estimated curvature ρ is calculated as a solution when the trajectory model evaluation function value J _ trj indicates an extreme value, and the trajectory model evaluation function value J _ trj includes a sum of a square sum error RSS of the past trajectory model Y 'coordinate value Ym' (k-i) and the past Y 'coordinate value Y' (k-i) and a product Wtrj · Pt _ J of the weighting coefficient and the trajectory interference function value. In this case, as is apparent from the reference expressions (14) to (16), the model expressions (21) and (22) described above include the estimated curvature ρ (═ 1/r), and therefore, by calculating the predicted X-coordinate value X (k + n) and the predicted Y-coordinate value Y (k + n) using this estimated curvature ρ, the predicted X-coordinate value X (k + n) and the predicted Y-coordinate value Y (k + n) can be calculated to values at which the degree of interference between the traffic participant and the interfering object becomes minimum and the error between the past trajectory model Y 'coordinate value Ym' (k-i) and the past Y 'coordinate value Y' (k-i) becomes minimum.

For the above reasons, it is possible to improve the calculation accuracy of the predicted speed V (k + n), the predicted X-coordinate value X (k + n), and the predicted Y-coordinate value Y (k + n), and to improve the prediction accuracy of the behavior of the traffic participant.

Further, as shown in the aforementioned expression (31), two X-coordinate-side modification coefficients Kx, Y-coordinate-side modification coefficient Ky are calculated from the squared sum error RSS, and the trajectory interference function value Pt _ j is calculated as a product of values obtained by modifying the X-coordinate-side reference value Pt _ X _ bs and the Y-coordinate-side reference value Pt _ Y _ bs with these values, and the velocity interference function value Pv _ j is also calculated by the same method. Therefore, in the case where the error between the past trajectory model Y 'coordinate value Ym' (k-i) and the past Y 'coordinate value Y' (k-i) is large and the degree of the past position trajectory of the traffic participant separating from the position trajectory model is large, it is possible to appropriately calculate the trajectory interference function value Pt _ j and the velocity interference function value Pv _ j while reflecting this.

Further, as the position trajectory model, a position trajectory model is used which is obtained by modeling the position trajectory of the traffic participant as an arc-shaped position trajectory in which a straight line extending in the traveling direction at the current point of the traffic participant is taken as a tangent line, and the predicted position (X (k + n), Y (k + n)) can be calculated with high accuracy based on this, unlike the case of patent document 1, even under the condition that the traffic participant moves while meandering or meandering.

In addition to this, since the automatic driving control can be executed using the predicted speed V (k + n) and the predicted position (X (k + n), Y (k + n)) of the traffic participant, which are calculated with high accuracy as described above, the control accuracy of the automatic driving can be improved.

In addition, since the speed disturbance function value Pv _ j and the trajectory disturbance function value Pt _ j are calculated in the disturbance function value calculation unit 8 independent of the ECU2, the calculation load of the ECU2 can be reduced.

In addition, the embodiment is an example in which a pedestrian is used as a traffic participant, and the traffic participant of the present invention is not limited to this, and may be any object as long as the object moves. For example, a vehicle, a mobile device, an animal, or the like may be a traffic participant.

In the embodiment, the predicted speed V (k + n), the predicted X-coordinate value X (k + n), and the predicted Y-coordinate value Y (k + n) are calculated as the action predicted value of the traffic participant, and only the predicted speed V (k + n) or only the predicted position (X (k + n), Y (k + n)) may be calculated as the action predicted value. When only the predicted speed V (k + n) is calculated, the speed disturbance function value Pv _ j may be calculated from the predicted speed V (k + n).

Further, the embodiment is an example in which a two-dimensional coordinate system model is used as the position trajectory model, but a three-dimensional coordinate system model may be used instead. In this case, the model value of the past position trajectory among the coordinate values of the remaining dimensions may be calculated using the three-dimensional coordinate system model and the one-dimensional coordinate value or the two-dimensional coordinate value of the past position trajectory of the traffic participant.

Further, the embodiment is an example in which the expression (23) is used as the trajectory model evaluation function value J _ trj, and instead, an evaluation function value in which the first term on the right side of the expression (23) (the square sum error RSS) is omitted and only the second term on the right side of the expression (23) (the sum of the multiplication values of the weighting coefficient and the interference function value) is used as a dependent variable may be used as the trajectory model evaluation function value J _ trj.

Further, in the embodiment, the expression (4) is used as the velocity model evaluation function value J _ v, but instead, the evaluation function value may be used by omitting the right first term of the expression (4) and using only the right second term (the sum of the multiplication values of the weighting coefficient and the interference function value) of the expression (4) as the dependent variable.

In addition, the embodiment is an example in which the error of the sum of squares of the past position trajectory and the past position trajectory model value is used as the error parameter, and the error parameter of the present invention is not limited to this as long as it indicates the error between the past position trajectory and the past position trajectory model value. For example, as the error parameter, an integrated value of a deviation between the past position trajectory and the past position trajectory model value may be used.

Further, the embodiment is an example in which a position trajectory model obtained by modeling the position trajectory of the traffic participant into an arc-shaped position trajectory in which a straight line extending in the traveling direction at the current point of the traffic participant is taken as a tangent line is used as the position trajectory model of the traffic participant, but the position trajectory model of the present invention is not limited to this, and may be any position trajectory model obtained by modeling the position trajectory of the traffic participant. For example, as the position trajectory model, a position trajectory model defined such that the movement trajectory of the traffic participant is changed by combining the line segment and the angle may be used.

Further, the embodiment is an example in which the behavior prediction device of the present invention is mounted on an autonomous vehicle, and the behavior prediction device of the present invention is not limited to this, and may be used alone or mounted on various devices such as a bicycle other than an autonomous vehicle.

On the other hand, the embodiment uses an example of a method of calculating the estimated acceleration α and the estimated curvature ρ as the behavior parameters so that the degree of interference between the traffic participant and the interfering object is minimized, but the method of calculating the interference degree parameter is not limited thereto, and may be any method as long as the estimated acceleration α and the estimated curvature ρ can be calculated so that the interference degree parameter is reduced.

Claims

1. A behavior prediction device for predicting the behavior of a traffic participant,

the disclosed device is provided with:

a surrounding situation recognition unit that recognizes surrounding situations of the traffic participant;

an action prediction value calculation unit that calculates an action prediction value that is a prediction value of a future action of the traffic participant, using a recognition result of the surrounding situation by the surrounding situation recognition unit and an action model that models an action of the traffic participant, the action model including an action mode parameter that indicates an action mode of the traffic participant;

an interference degree parameter calculation unit that calculates an interference degree parameter indicating an interference degree with the traffic participant of an interference object around the traffic participant, using the action prediction value; and

and an action mode parameter determining unit configured to determine the action mode parameter such that the interference level indicated by the interference level parameter is reduced.

2. The action prediction device according to claim 1,

the action of the traffic participant is a time-series position trajectory representing the actual spatial position of the traffic participant,

the action model is a location trajectory model that models the location trajectory of the traffic participant,

the action prediction value is a prediction value of the location trajectory,

the behavior prediction device further includes:

a position trajectory acquisition unit that acquires the position trajectory of the traffic participant;

a past position trajectory storage unit that stores a past position trajectory that is a past value of the position trajectory acquired by the position trajectory acquisition unit;

a past position trajectory model value calculation means for calculating a past position trajectory model value as a model value of the past position trajectory using the behavior pattern parameter and a coordinate value as at least one component in the past position trajectory stored in the past trajectory storage means; and

an error parameter calculation unit that calculates an error parameter indicating an error between the past position trajectory and the past position trajectory model value,

the behavior parameter determination unit determines the behavior parameter such that an error indicated by the error parameter is further reduced in addition to the degree of interference indicated by the degree of interference parameter.

3. The action prediction device according to claim 2,

the position trajectory is a position trajectory of a two-dimensional coordinate system having the first coordinate value and the second coordinate value as components,

in the past position trajectory, a first coordinate past value as a past value of the first coordinate value and a second coordinate past value as a past value of the second coordinate value are taken as components,

the past position trajectory model value calculation unit calculates a second coordinate past model value as a model value of the second coordinate past value as the past position trajectory model value using the behavior pattern parameter and the first coordinate past value in the past position trajectory,

the error parameter calculation unit calculates a value representing an error between the second coordinate past value and the second coordinate past model value as the error parameter.

4. The action prediction device according to claim 2 or 3,

the interference degree parameter calculation unit calculates a modification value according to the error parameter, and calculates the interference degree parameter using a value obtained by modifying the reference value of the interference degree parameter with the modification value.

5. The action prediction device according to any one of claims 2 to 4,

the position trajectory model is a position trajectory model obtained by modeling the position trajectory of the traffic participant as an arc-shaped position trajectory in which a straight line extending in a traveling direction of a current point of the traffic participant is defined as a tangent line.

6. The action prediction device according to any one of claims 2 to 5,

the action mode parameter determination unit determines the action mode parameter as a solution when the evaluation function represents an extreme value, using an evaluation function including the disturbance degree parameter and the error parameter as arguments.

7. The action prediction device according to claim 1,

the action mode parameter determination unit determines the action mode parameter as a solution when the evaluation function represents an extremum using an evaluation function including the disturbance degree parameter as an argument.

8. The action prediction device according to claim 1,

the action of the traffic participant is at least one of a position trajectory representing a time series of actual spatial positions of the traffic participant and a speed of the traffic participant.

9. An automatic driving device is characterized in that,

the disclosed device is provided with:

the action prediction device of any one of claims 1 to 8; and

and a control unit that executes automatic driving control of the vehicle based on the action prediction value.