CN-121809287-B - Automatic driving collision dangerous scene generation method based on reverse motion reconstruction

CN121809287BCN 121809287 BCN121809287 BCN 121809287BCN-121809287-B

Abstract

The invention belongs to the technical field of automatic driving automobile testing, and particularly relates to an automatic driving collision dangerous scene generation method based on reverse motion reconstruction. Firstly, modeling a vehicle running state, modeling a vehicle reverse motion process into a mathematical expression form, then establishing a starting point steady running state set, and defining steady state constraint, thirdly, defining sampling process constraint, introducing vehicle dynamics and road models, avoiding that the vehicle motion does not accord with physical constraint, and fourthly, converting a generated interaction track into a scene description form. The method can help enterprises establish a set of efficient collision dangerous scene generation system, improves the efficiency of dangerous scene generation, reduces the waste of calculation power, finally establishes a complete dangerous scene database and accelerates the test and verification of the automatic driving automobile.

Inventors

ZHANG PEIXING
Kou Hongrui
ZHU BING
ZHAO JIAN
HAN JIAYI

Assignees

吉林大学

Dates

Publication Date: 20260508
Application Date: 20260306

Claims (4)

1. The automatic driving collision dangerous scene generation method based on reverse motion reconstruction is characterized by comprising the following steps of: modeling a vehicle running state, and modeling a vehicle reverse motion process into a mathematical expression form, wherein the single vehicle running state is analyzed, and the single vehicle running state is converted into the mathematical description form; setting the occurrence time of traffic collision event At the moment, given the relative pose, speed and heading state of the involved collision vehicles and the geometric information of the road and the lane, the finally generated time range of the track set before collision meeting the vehicle dynamics and the road constraint is The track is at The moment is consistent with the given final state constraint of collision, in When the device is in a stable running state; Firstly, carrying out mathematical description on the motion state of the vehicle, describing various states of the vehicle by using a road Frenet coordinate system, wherein the state vector is defined as: ; in the formula, Is the longitudinal position; is laterally offset; is heading error; Is the speed; Is in a transverse and posture movement state; subsequently, the key information in the track generation process of the vehicle is the control behavior of the vehicle, including steering and braking, and the control input to the vehicle is expressed as: ; in the formula, Is steering control; is the longitudinal acceleration; The position state of the vehicle at each moment is expressed as: ; in the formula, For standard vehicle dynamics and road geometry A state transfer function under the combined action; Is that A vehicle state vector for a moment; Is that A vehicle control input of time; Is that Road geometry at time; step two, establishing a starting point steady running state set, and defining steady state constraint; step three, in order to define the constraint of the sampling process, introducing a vehicle dynamics and road model to avoid that the vehicle motion does not accord with the physical constraint, the specific method is as follows: s31, converting the preamble behavior of the collision state into a mathematical expression form: ; in the formula, Is that A state vector of the vehicle a at the moment; Is that A state vector of the vehicle B at the moment; Is that Control behavior of vehicle a from time to time 0; Is that Control behavior of vehicle B from time to time 0; ; To meet the state set of the vehicle conforming to the physical rule, the track evolution meets the dynamic constraint defined in S12 The relative state must be Is a set of collisions satisfying the definition in S13 ; S32, according to S31, for each sample Candidate trajectories obtained by forward integration are: ; for each sample, the constraint optimization problem is solved Correcting the obtained candidate track: ; ; in the formula, Is the first Control sequence of vehicle in whole section Is a cost function of (2); Is that Time of day (time) A state vector of the vehicle; Is that Time of day (time) A control input vector for the vehicle; a vehicle steady state set conforming to a physical rule; Solving through a sampling process to obtain a series of running tracks meeting collision states, wherein the initial state is stable running, the end point is collided, and meanwhile, the state change process of each step meets the vehicle dynamics requirement; and step four, converting the generated interaction track into a scene description form.
2. The method for generating an automatic driving collision risk scene based on reverse motion reconstruction according to claim 1, wherein the first step further comprises the steps of: analyzing the interaction motion state of the two vehicles, and converting the interaction motion state of the two vehicles into a mathematical description form; Let the states of the vehicle A and the vehicle B be respectively And (3) with At this time, a relative state vector is introduced Describing the interaction process between two vehicles: ; in the formula, Constructing variables of relative longitudinal distance, lateral offset, relative heading and relative speed between vehicles; since vehicles A and B each satisfy the dynamics constraint The relative state at this time satisfies: ; in the formula, The dynamic difference of the double vehicles is determined together with the road geometry; Is that The relative states of vehicles A and B at the moment; Is that Control input of the vehicle A at the moment; Is that Control input of the vehicle B at the moment; According to the interactive motion state of the vehicle, carrying out mathematical form description on the collision state set; The collision state is represented by a set in the relative motion state space as: ; in the formula, The minimum distance function of the geometric envelope of the two vehicles; extracting a collision-related relative quantity; for a given collision relative state; the relative motion state between different vehicles at the initial moment; A collision-related relative quantity between different vehicles extracted for an initial time; A collision-related relative amount threshold value set; generating a plurality of collision interaction tracks for a single collision state given the single collision data.
3. The method for generating the collision risk scene of the automatic driving based on the reverse motion reconstruction according to claim 1, wherein the specific method of the second step is as follows: s21, starting the track Limiting to a stable driving state and serving as a hard constraint in the generation process; The steady running state set is defined as: ; in the formula, , , , , , , The steady driving parameter boundaries obtained by using normal distribution for counting natural driving data are respectively a transverse deviation maximum value, a course deviation maximum value, a speed minimum value, a speed maximum value, a steering angle maximum value, a control input maximum value and a control input change maximum value; Is that A vehicle state vector for a moment; to control the amount of input variation.
4. The method for generating the collision risk scene of the automatic driving based on the reverse motion reconstruction according to claim 1, wherein the specific method of the fourth step is as follows: s41, completing other scene elements except road shapes and vehicle motion interaction according to the six-layer scene model, collecting the other scene elements and the current acquired track state by a combined test method to obtain collision danger scene data for testing, and constructing a test scene library.

Description

Automatic driving collision dangerous scene generation method based on reverse motion reconstruction Technical Field The invention belongs to the technical field of automatic driving automobile testing, and particularly relates to an automatic driving collision dangerous scene generation method based on reverse motion reconstruction. Background The collision risk scene is a key content in the test process of the automatic driving automobile, and in order to find the performance defect of the automatic driving automobile, a test scene library covering a large number of collision risk scenes needs to be constructed. Currently, the automatic driving automobile collision dangerous scene mostly adopts a forward generation mode, namely, the initial running state of the stable running of a given automobile, and the motion behavior of the automobile is adjusted step by step through reinforcement learning, optimized searching and other modes, so that the occurrence of a final collision accident is induced. However, since the sampling interval is very short, a great deal of behavior selection is needed in the scene generation process in theory, and since collision is a small probability time, most of behavior combination results are safety scenes in the training process, so that the calculation force is wasted. How to reduce the safety interaction in the vehicle behavior interaction, and ensure that each generated scene is the dangerous interaction which is possibly collided, is the key point of the current collision dangerous scene generation. Disclosure of Invention In order to solve the problems, the invention provides an automatic driving collision dangerous scene generation method based on reverse motion reconstruction, which is characterized in that a vehicle collision is taken as a sampling starting point, the possible vehicle behavior at the previous moment is continuously deduced through a given vehicle dynamics behavior constraint in a reverse sampling mode until the vehicle is sampled to a stable running state. Firstly, modeling a vehicle running state, modeling a vehicle reverse motion process into a mathematical expression form, then establishing a starting point steady running state set, and defining steady state constraint, thirdly, defining sampling process constraint, introducing vehicle dynamics and road models, avoiding that the vehicle motion does not accord with physical constraint, and fourthly, converting a generated interaction track into a scene description form. The method can help enterprises establish a set of efficient collision dangerous scene generation system, improves the efficiency of dangerous scene generation, reduces the waste of calculation power, finally establishes a complete dangerous scene database and accelerates the test and verification of the automatic driving automobile. The technical scheme of the invention is as follows in combination with the accompanying drawings: The invention provides an automatic driving collision dangerous scene generation method based on reverse motion reconstruction, which comprises the following steps of: firstly, modeling a vehicle running state, and modeling a vehicle reverse motion process into a mathematical expression form; step two, establishing a starting point steady running state set, and defining steady state constraint; Step three, introducing vehicle dynamics and a road model to ensure the constraint of the sampling process, so as to avoid that the vehicle motion does not accord with the physical constraint; and step four, converting the generated interaction track into a scene description form. Further, the specific method of the first step is as follows: s11, analyzing the bicycle motion state, and converting the bicycle motion state into a mathematical description form; setting the occurrence time of traffic collision event At the moment, given the relative pose, speed and heading state of the involved collision vehicles and the geometric information of the road and the lane, the finally generated time range of the track set before collision meeting the vehicle dynamics and the road constraint isThe track is atThe moment is consistent with the given final state constraint of collision, inWhen the device is in a stable running state; Firstly, carrying out mathematical description on the motion state of the vehicle, describing various states of the vehicle by using a road Frenet coordinate system, wherein the state vector is defined as: ; in the formula, Is the longitudinal position; is laterally offset; is heading error; Is the speed; Is in a transverse and posture movement state; subsequently, the key information in the track generation process of the vehicle is the control behavior of the vehicle, including steering and braking, and the control input to the vehicle is expressed as: ; in the formula, Is steering control; is the longitudinal acceleration; The position state of the vehicle at each moment is expressed as: ; in t