An automatic driving collision danger scene generation method based on inverse motion reconstruction

Abstract

The present application belongs to the technical field of automatic driving car testing, and in particular to an automatic driving collision danger scene generation method based on reverse motion reconstruction. First, the vehicle operating state is modeled, and the reverse motion process of the vehicle is modeled into a mathematical expression form. Then, a starting point stable driving state set is established, and the smooth state constraint is clarified. The third step is to clarify the sampling process constraint, introduce the vehicle dynamics and road model, and avoid the vehicle motion not conforming to the physical constraint. The fourth step is to convert the generated interaction trajectory into a scene description form. The present application can help enterprises establish an efficient collision danger scene generation system, improve the efficiency of danger scene generation, reduce the waste of computing power, and ultimately establish a complete danger scene database to accelerate the testing and verification of automatic driving cars.

Description

Technical Field

[0001] This invention belongs to the field of autonomous vehicle testing technology, specifically a method for generating autonomous driving collision hazard scenarios based on reverse motion reconstruction. Background Technology

[0002] Collision hazard scenarios are a crucial aspect of autonomous vehicle testing. To identify performance defects in autonomous vehicles, a comprehensive test scenario library covering numerous collision hazard scenarios is needed. Currently, most autonomous vehicle collision hazard scenarios are generated using a forward generation approach. This involves giving the vehicle an initial stable driving state and then gradually adjusting its behavior through reinforcement learning and optimization search to induce a collision. However, due to the short sampling intervals, theoretically, a single scenario generation process requires a large number of behavior selections. Furthermore, since collisions are low-probability events, most behavior combinations during training result in safe scenarios, leading to wasted computational resources. The key focus of current collision hazard scenario generation is to reduce safe interactions in vehicle behavior and ensure that each generated scenario represents a potentially dangerous interaction that could lead to a collision. Summary of the Invention

[0003] To address the aforementioned issues, this invention provides a method for generating autonomous driving collision hazard scenarios based on inverse motion reconstruction. The core of this method lies in using vehicle collision as the sampling starting point, and given vehicle dynamics constraints, continuously inferring possible vehicle behaviors from the previous moment through inverse sampling until a stable vehicle operating state is sampled. First, vehicle operating state modeling is performed, modeling the vehicle's inverse motion process into a mathematical expression; then, a set of stable driving states at the starting point is established, clarifying the constraints of the stable state; the third step is to clarify the sampling process constraints by introducing vehicle dynamics and road models to avoid vehicle motion not conforming to physical constraints; the fourth step is to convert the generated interaction trajectory into a scene description. The results of this invention can help enterprises establish an efficient collision hazard scenario generation system, improve the efficiency of hazard scenario generation, reduce computational waste, and ultimately establish a complete hazard scenario database, accelerating the testing and verification of autonomous vehicles.

[0004] The technical solution of this invention is described below in conjunction with the accompanying drawings:

[0005] This invention provides a method for generating autonomous driving collision hazard scenarios based on inverse motion reconstruction, comprising the following steps:

[0006] Step 1: Model the vehicle's operating state, and express the vehicle's reverse motion process in a mathematical form;

[0007] Step 2: Establish a set of stable driving states at the starting point and define the constraints of stable states;

[0008] Step 3: To clarify the constraints of the sampling process, vehicle dynamics and road models are introduced to avoid vehicle motion that does not conform to physical constraints;

[0009] Step 4: Convert the generated interaction trajectory into a scene description.

[0010] Furthermore, the specific method for step one is as follows:

[0011] S11. Analyze the motion state of the bicycle and transform it into a mathematical description.

[0012] Assume the traffic collision occurs at time [time]. At this point, given the relative pose, velocity, and heading of the vehicles involved in the collision, as well as the geometric information of the road and lanes, the final generated set of pre-collision trajectories that satisfies vehicle dynamics and road constraints has a time range of [missing information]. ; trajectory in The time coincides with the given collision final state constraint, in At that time, it is in a stable operating state;

[0013] First, the vehicle's motion state is mathematically described; the Frenet coordinate system is used to describe various vehicle states, and the state vector is defined as follows:

[0014] ;

[0015] In the formula, Vertical position; This is a horizontal offset; For heading error; For speed; This refers to lateral and postureal motion states;

[0016] Subsequently, the key information in the vehicle trajectory generation process is the vehicle's control behavior, including steering and braking. The control inputs to the vehicle are represented as follows:

[0017] ;

[0018] In the formula, For steering control; It is longitudinal acceleration;

[0019] The vehicle's position at each moment is represented as:

[0020] ;

[0021] In the formula, For standard vehicle dynamics and road geometry State transition function under combined action; for The vehicle state vector at time t; for Vehicle control input at any given time; for The road geometry at any given time;

[0022] S12. Analyze the interactive motion state of the two vehicles and transform the interactive motion state of the two vehicles into a mathematical description.

[0023] Let the states of vehicle A and vehicle B be respectively... and At this point, a relative state vector is introduced. Describe the interaction process between the two vehicles:

[0024] ;

[0025] In the formula, The variables of relative longitudinal distance, lateral offset, relative heading, and relative speed between vehicles are constructed.

[0026] Since vehicles A and B each satisfy dynamic constraints At this point, the relative state satisfies:

[0027] ;

[0028] In the formula, It is determined by both the dual-vehicle dynamics difference and the road geometry; for The relative states of vehicles A and B at any given time; for Control input for vehicle A at any given time; for Control input for vehicle B at any given time;

[0029] S13. Based on the vehicle's interactive motion state, describe the collision state set in a mathematical form;

[0030] Collision states are represented by a set in the relative motion state space as follows:

[0031] ;

[0032] In the formula, Let be the minimum distance function of the geometric envelopes of the two vehicles; Extract collision-related relative quantities; Given the relative states of the collision; This represents the relative motion state between the different vehicles at the initial moment; The relative collision-related quantities between different vehicles extracted at the initial moment; The threshold value for the relative quantity related to the collision is set.

[0033] Given a single collision data point, multiple collision interaction trajectories are generated for a single collision state.

[0034] Furthermore, the specific method for step two is as follows:

[0035] S21, Set the trajectory starting point It is restricted to a stable driving state and serves as a hard constraint in the generation process;

[0036] The set of stable driving states is defined as follows:

[0037] ;

[0038] In the formula, , , , , , , The boundaries of steady-state driving parameters obtained by using the normal distribution for statistical analysis of natural driving data are: maximum lateral offset, maximum heading deviation, minimum speed, maximum speed, maximum steering angle, maximum control input, and maximum control input change. for The vehicle state vector at time t; To control the amount of input variation;

[0039] Furthermore, the specific method for step three is as follows:

[0040] S31. Transform the preceding behavior of the collision state into a mathematical expression:

[0041] ;

[0042] In the formula, for The state vector of vehicle A at time t; for The state vector of vehicle B at time t; for Control behavior of vehicle A from time 0 to time 0; for Control behavior of vehicle B from time 0 to time 0; ; The vehicle state set conforms to physical laws; the trajectory evolution satisfies the dynamic constraints defined in S12. The relative state must be It is a collision set that satisfies the definition in S13. ;

[0043] S32. According to S31, for each sample The candidate trajectory obtained by forward integration is:

[0044] ;

[0045] For each sample, solve the following constrained optimization problem. The obtained candidate trajectories are then corrected:

[0046] ;

[0047] ;

[0048] In the formula, For the first The vehicle in the entire control sequence The cost function; for Time of the first The vehicle's state vector; for Time of the first The vehicle's control input vector; The set of vehicle steady-state states that conform to physical laws;

[0049] By solving through a sampling process, a series of driving trajectories that satisfy the collision state are obtained. The initial state is stable operation, and the end point is a collision. At the same time, each step of the state change process satisfies the vehicle dynamics requirements.

[0050] Furthermore, the specific method for step four is as follows:

[0051] S41. Based on the six-layer scene model, complete the scene elements other than road shape and vehicle motion interaction. Combine the other scene elements with the currently acquired trajectory state through the combination test method to obtain collision hazard scene data for testing and build a test scene library.

[0052] The beneficial effects of this invention are as follows:

[0053] The results of this invention can help enterprises establish an efficient collision hazard scenario generation system, improve the efficiency of hazard scenario generation, reduce computing power waste, and ultimately establish a complete hazard scenario database to accelerate the testing and verification of autonomous vehicles. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0056] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0057] See Figure 1 This invention provides a method for generating autonomous driving collision hazard scenarios based on inverse motion reconstruction, characterized by the following steps:

[0058] Step 1: Model the vehicle's operating state, transforming the vehicle's reverse motion process into a mathematical expression. The specific method is as follows:

[0059] S11. Analyze the motion state of the bicycle and transform it into a mathematical description.

[0060] Assume the traffic collision occurs at time [time]. At this point, given the relative pose, velocity, and heading of the vehicles involved in the collision, as well as the geometric information of the road and lanes, the final generated set of pre-collision trajectories that satisfies vehicle dynamics and road constraints has a time range of [missing information]. ; trajectory in The time coincides with the given collision final state constraint, in At this time, the vehicle is in a stable operating state, meaning that the vehicle is within the lane, the attitude and control changes are gradual, and there are no significant dangerous interactions.

[0061] First, the vehicle's motion state is mathematically described; the Frenet coordinate system is used to describe various vehicle states, and the state vector is defined as follows:

[0062] ;

[0063] In the formula, Vertical position; This is a horizontal offset; For heading error; For speed; This refers to lateral and attitudinal motion states (such as sideslip angle, yaw rate, etc.).

[0064] Subsequently, the key information in the vehicle trajectory generation process is the vehicle's control behavior, including steering and braking. The control inputs to the vehicle are represented as follows:

[0065] ;

[0066] In the formula, For steering control; It is longitudinal acceleration;

[0067] The vehicle's position at each moment is represented as:

[0068] ;

[0069] In the formula, For standard vehicle dynamics and road geometry State transition function under combined action; for The vehicle state vector at time t; for Vehicle control input at any given time; for The road geometry at any given time;

[0070] S12. Analyze the interactive motion state of the two vehicles and transform the interactive motion state of the two vehicles into a mathematical description.

[0071] Let the states of vehicle A and vehicle B be respectively... and At this point, a relative state vector is introduced. Describe the interaction process between the two vehicles:

[0072] ;

[0073] In the formula, The variables of relative longitudinal distance, lateral offset, relative heading, and relative speed between vehicles are constructed.

[0074] Since vehicles A and B each satisfy dynamic constraints At this point, the relative state satisfies:

[0075] ;

[0076] In the formula, It is determined by both the dual-vehicle dynamics difference and the road geometry; for The relative states of vehicles A and B at any given time; for Control input for vehicle A at any given time; for Control input for vehicle B at any given time;

[0077] S13. Based on the vehicle's interactive motion state, describe the collision state set in mathematical form.

[0078] Collision states are represented by a set in the relative motion state space as follows:

[0079] ;

[0080] In the formula, Let be the minimum distance function of the geometric envelopes of the two vehicles; Extract collision-related relative quantities; Given the relative states of the collision; This represents the relative motion state between the different vehicles at the initial moment; The relative collision-related quantities between different vehicles extracted at the initial moment; The threshold value for the relative quantity related to the collision is set.

[0081] Given a single collision data point, multiple collision interaction trajectories are generated for a single collision state.

[0082] Step 2: Establish a set of stable driving states at the starting point and define the constraints of stable states. The specific method is as follows:

[0083] S21, Set the trajectory starting point It is restricted to a stable driving state and serves as a hard constraint in the generation process;

[0084] The set of stable driving states is defined as follows:

[0085] ;

[0086] In the formula, , , , , , , The boundaries of steady-state driving parameters obtained by using the normal distribution for statistical analysis of natural driving data are: maximum lateral offset, maximum heading deviation, minimum speed, maximum speed, maximum steering angle, maximum control input, and maximum control input change. for The vehicle state vector at time t; To control the amount of input variation;

[0087] Step 3: To clarify the constraints of the sampling process, vehicle dynamics and road models are introduced to avoid vehicle motion from violating physical constraints. The specific method is as follows:

[0088] S31. Transform the preceding behavior of the collision state into a mathematical expression:

[0089] ;

[0090] In the formula, for The state vector of vehicle A at time t; for The state vector of vehicle B at time t; for Control behavior of vehicle A from time 0 to time 0; for Control behavior of vehicle B from time 0 to time 0; ; The vehicle state set conforms to physical laws; the trajectory evolution satisfies the dynamic constraints defined in S12. The relative state must be It is a collision set that satisfies the definition in S13. ;

[0091] S32. According to S31, for each sample The candidate trajectory obtained by forward integration is:

[0092] ;

[0093] For each sample, solve the following constrained optimization problem. The obtained candidate trajectories are then corrected:

[0094] ;

[0095] ;

[0096] In the formula, For the first The vehicle in the entire control sequence The cost function; for Time of the first The vehicle's state vector; for Time of the first The vehicle's control input vector; The set of vehicle steady-state states that conform to physical laws;

[0097] By solving through a sampling process, a series of driving trajectories that satisfy the collision state are obtained. The initial state is stable operation, and the end point is a collision. At the same time, each step of the state change process satisfies the vehicle dynamics requirements.

[0098] Step 4: Convert the generated interaction trajectory into a scene description. The specific method is as follows:

[0099] S41. Based on the six-layer scene model, complete the scene elements other than road shape and vehicle motion interaction. Combine the other scene elements with the currently acquired trajectory state through the combination test method to obtain collision hazard scene data for testing and build a test scene library.

[0100] In summary, this invention can help enterprises establish an efficient collision hazard scenario generation system, improve the efficiency of hazard scenario generation, reduce computing power waste, and ultimately establish a complete hazard scenario database to accelerate the testing and verification of autonomous vehicles.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for generating autonomous driving collision hazard scenarios based on inverse motion reconstruction, characterized in that, Includes the following steps: Step 1: Model the vehicle's operating state and transform the vehicle's reverse motion process into a mathematical expression. This involves analyzing the motion state of a single vehicle and converting it into a mathematical description. Assume the traffic collision occurs at time [time]. At this point, given the relative pose, velocity, and heading of the vehicles involved in the collision, as well as the geometric information of the road and lanes, the final generated set of pre-collision trajectories that satisfies vehicle dynamics and road constraints has a time range of [missing information]. ; trajectory in The time coincides with the given collision final state constraint, in At that time, it is in a stable operating state; First, the vehicle's motion state is mathematically described; the Frenet coordinate system is used to describe various vehicle states, and the state vector is defined as follows: ； In the formula, Vertical position; This is a horizontal offset; For heading error; For speed; This refers to lateral and postureal motion states; Subsequently, the key information in the vehicle trajectory generation process is the vehicle's control behavior, including steering and braking. The control inputs for the vehicle are represented as follows: ； In the formula, For steering control; It is longitudinal acceleration; The vehicle's position at each moment is represented as: ； In the formula, For standard vehicle dynamics and road geometry State transition function under combined action; for The vehicle state vector at time t; for Vehicle control input at any given time; for The road geometry at any given time; Step 2: Establish a set of stable driving states at the starting point and define the constraints of stable states; Step 3: To clarify the constraints of the sampling process, vehicle dynamics and road models are introduced to avoid vehicle motion from violating physical constraints. The specific method is as follows: S31. Transform the preceding behavior of the collision state into a mathematical expression: ； In the formula, for The state vector of vehicle A at time t; for The state vector of vehicle B at time t; for Control behavior of vehicle A from time 0 to time 0; for Control behavior of vehicle B from time 0 to time 0; ; The vehicle state set conforms to physical laws; the trajectory evolution satisfies the dynamic constraints defined in S12. The relative state must be It is a collision set that satisfies the definition in S13. ; S32. According to S31, for each sample The candidate trajectory obtained by forward integration is: ； For each sample, solve the following constrained optimization problem. The obtained candidate trajectories are then corrected: ； ； In the formula, For the first The vehicle in the entire control sequence The cost function; for Time of the first The vehicle's state vector; for Time of the first The vehicle's control input vector; The set of vehicle steady-state states that conform to physical laws; By solving through the sampling process, a series of driving trajectories that satisfy the collision state are obtained. The initial state is stable operation, and the end point is a collision. At the same time, each step of the state change process satisfies the vehicle dynamics requirements. Step 4: Convert the generated interaction trajectory into a scene description.

2. The method for generating autonomous driving collision hazard scenarios based on inverse motion reconstruction according to claim 1, characterized in that, Step one also includes the following steps: Analyze the interactive motion states of the two vehicles and transform them into a mathematical description. Let the states of vehicle A and vehicle B be respectively... and At this point, a relative state vector is introduced. Describe the interaction process between the two vehicles: ； In the formula, The variables of relative longitudinal distance, lateral offset, relative heading, and relative speed between vehicles are constructed. Since vehicles A and B each satisfy dynamic constraints At this point, the relative state satisfies: ； In the formula, It is determined by both the dual-vehicle dynamics difference and the road geometry; for The relative states of vehicles A and B at any given time; for Control input for vehicle A at any given time; for Control input for vehicle B at any given time; Based on the vehicle's interactive motion state, the collision state set is described in mathematical form; Collision states are represented by a set in the relative motion state space as follows: ； In the formula, Let be the minimum distance function of the geometric envelopes of the two vehicles; Extract collision-related relative quantities; Given the relative states of the collision; This represents the relative motion state between the different vehicles at the initial moment; The relative collision-related quantities between different vehicles extracted at the initial moment; The threshold value for the relative quantity related to the collision is set. Given a single collision data point, multiple collision interaction trajectories are generated for a single collision state.

3. The method for generating autonomous driving collision hazard scenarios based on inverse motion reconstruction according to claim 1, characterized in that, The specific method for step two is as follows: S21, Set the starting point of the trajectory It is restricted to a stable driving state and serves as a hard constraint in the generation process; The set of stable driving states is defined as follows: ； In the formula, , , , , , , The steady-state driving parameter boundaries obtained using the normal distribution for statistical analysis of natural driving data are: maximum lateral offset, maximum heading deviation, minimum speed, maximum speed, maximum steering angle, maximum control input, and maximum control input change. for The vehicle state vector at time t; To control the amount of input variation.

4. The method for generating autonomous driving collision hazard scenarios based on inverse motion reconstruction according to claim 1, characterized in that, The specific method for step four is as follows: S41. Based on the six-layer scene model, complete the scene elements other than road shape and vehicle motion interaction. Combine the other scene elements with the currently acquired trajectory state through the combination test method to obtain collision hazard scene data for testing and build a test scene library.

Images