An accident report-based automatic driving generalization test scene construction method

Abstract

The present application relates to an automatic driving generalization test scene construction method based on accident reports. According to the layer-based scene definition, the accident scene is initially abstracted into three layers: road network and traffic guiding objects, environmental conditions and dynamic objects. Then, the LLM is used to systematically extract information about the accident influencing factors from the accident reports, and the information is organized layer by layer. Subsequently, constraints on the pre-accident driving behavior of traffic participants are established, and a constraint solver is used to generate trajectories that meet the specified constraints in the accident report. Then, the extracted environmental and road information and the generated trajectories are input into the driving simulator to reconstruct the accident scene. Finally, in the evaluation stage, the generated simulation scene is converted into test cases containing test oracles, and these test cases are subjected to automatic driving simulation testing. The present application aims to solve the problem of reproducing accident scenes on different roads, thereby helping automatic driving system practitioners to carry out testing work in accident scenes with different road structures.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of safety testing of autonomous driving systems (ADS), and is particularly suitable for the field of generalization testing scenario construction of autonomous driving. The purpose is to provide guidance for ADS practitioners to carry out testing work under different road structure accident scenarios, and is a technology for automatically generating road general scenarios from accident reports. BACKGROUND

[0002] The advent of autonomous driving technology has opened a new era of transportation, promising to improve safety, efficiency, and convenience. However, accidents involving autonomous vehicles, including those resulting in death, have proven that autonomous driving is not as safe as it claims. In many cases, these collisions can be attributed to defective software, highlighting the urgent need for enhanced methods of testing autonomous driving software.

[0003] ADS testing, as one of the key quality assurance techniques, has received extensive attention from academia and industry. Key testing techniques for ADS can be divided into two categories: road testing and simulation testing. Road testing includes testing specific driving scenarios in a closed autonomous vehicle test field or monitoring autonomous vehicles in real traffic, but this method requires a long time and a large amount of resources. In addition, obtaining various key testing data, including a wide range of actual use scenarios, poses a major challenge to testers. Therefore, high-fidelity simulation testing methods have become a necessary condition for the development and verification of autonomous driving technology. ADS testing on simulation platforms such as LGSVL, CARLA, etc. An important aspect of ADS simulation testing is to identify and construct key scenarios that can lead to accidents.

[0004] Software engineering researchers have proposed using real-life accident cases to generate key testing scenarios, as real car accidents pose a significant challenge to autonomous vehicles. Therefore, recent research has focused on scenario reconstruction from different driving data sources. This includes using text descriptions, accident sketches, sensor data, and video recordings to generate basic scenarios. Compared with sensor data and video recordings of key cases, text descriptions of collisions are easier to obtain and more abundant. Accident report analysis methods provide more comprehensive collision insights, including complex details such as weather, lighting, and road conditions, which are difficult to represent visually. However, current scenario reconstruction methods based on accident reports have limited accuracy in extracting information. In addition, current methods can only reproduce accident scenarios on the same road structure described in the accident report. Since the road structure in the simulation may be different from the road structure described in the accident report, the application is limited to simulation-based ADS testing. Reproducing accident scenarios on different roads is a challenging task.

[0005] To fill this gap, we propose an automatic and general method to reconstruct accident scenarios on different road structures from accident reports. The main goal of this method is to automatically reconstruct accident scenarios from accident reports on different roads in a simulated environment, which are then used as initial seeds for ADS testing. This method first uses a carefully designed language prompt pattern to guide the LLM to accurately extract the environment, road, and object motion information. Based on the extracted information and road reconstruction scenarios, this method formulates constraints for accident scenarios. By solving these constraints with a constraint solver, the accident trajectories of the accident participants are generated, and the accident scenario is finally reconstructed. Therefore, the generated scenarios can be executed on the required map. In addition, in order to use these scenarios for ADS testing, this method converts them into test scenarios by identifying the ego vehicle (i.e., the vehicle controlled by the ADS in the test) and NPC (non-player character) vehicles. The NPC vehicles will follow the calculated trajectories, while the ego vehicle is controlled by the ADS, rather than following a predetermined trajectory. SUMMARY

[0006] The present application effectively solves the problem that the test strength requirement of new version software variation is unknown by providing a technology for automatically generating road general scenarios from accident reports, thereby helping testers to automatically generate road general scenarios and helping testers to carry out test work in accident scenarios of different road structures.

[0007] To achieve the above-mentioned goal, the present application proposes an automatic driving generalization test scenario construction method based on accident reports, characterized in that according to the layer-based scene definition, the accident scenario information is initially abstracted into three layers: road network and traffic guiding objects, environmental conditions and dynamic objects. Then, a large language model (LLM) is used to systematically extract information about accident influencing factors from accident reports, and the information is organized layer by layer. Subsequently, constraints are established for the pre-accident driving behavior of traffic participants, and a constraint solver is used to generate trajectories that meet the specified constraints in the accident report. Then, the extracted environmental and road information trajectories are input into a driving simulator to reconstruct the accident scene. Finally, in the evaluation stage, the generated simulation scenarios are converted into test cases containing test oracles, and these test cases are input into the automatic driving system (ADS). This method checks whether the tested ADS successfully reaches the final expected position without encountering any crashes. The method includes the following steps:

[0008] 1) Information extraction; To abstractly represent the accident information and organize it into a semantic structure that the LLM can understand, the extracted information is presented in a hierarchical representation, divided into three layers: environment, road, and dynamic objects. The environment layer includes weather and lighting conditions. Weather conditions include rain, fog, snow, and other factors. Lighting conditions are related to the lighting conditions on the road, which are usually brighter during the day and darker at night. In addition, streetlights can enhance lighting. The road layer describes the road network and traffic signs used for road guidance. Road represents the geographical context of the collision, including the type of road where the accident occurred (Collision Location) and the number of lanes on the related road (LaneNum). In addition, the method extracts information about the speed limit on the road (SpeedLimit) to reconstruct the speed constraints applicable to the accident location. The dynamic objects layer contains information about the impactor and the victim, as well as the moving actions that led to the collision. The method extracts information about the number of participants involved in the simultaneous collision (ParticipantsNumber) and identifies the type of collision. CrashType information specifies the angle at which the traffic participants collided, including three types of collisions: rear-end collision, head-on collision, and front-back collision. For each vehicle involved in the accident, the method extracts the state of each vehicle, including the initial running lane of the vehicle (RunningLanes), the initial driving direction of the vehicle (DrivingDirections), and the behavior of the vehicle before the collision (DrivingActions). Specifically, vehicle behavior describes the normal and abnormal behavior of the vehicle. Normal driving actions include U-turn, parking, entering the road, vehicle crossing, left turn, right turn, following lane, and lane change. Abnormal driving behavior includes driving off the road and driving in reverse.

[0009] The method uses a prompt pattern to extract accident information layer by layer. For each language pattern, the meaning of each attribute is first explained to help the LLM understand the extracted information. In addition, the pattern also includes heuristic rules to guide the LLM to produce accurate results. Finally, few-shot learning is used to ensure that the output of the LLM meets the expected standards.

[0010] 2) Trajectory planning; this method uses the extracted driving actions and road information to simulate the collision and calculate the trajectories of simulated traffic participants. The road information R extracted in step 1), the dynamic object information D extracted in step 1), the given map MAP, and the defined driving action constraints C are input. First, the given map is parsed into a set of candidate roads, and the road type of each road (e.g., intersection) is noted. Then the candidate road set is iterated, and lanes of different lengths and widths are selected in turn. If the current road type matches the road type where the accident occurred, and the maximum number of lanes occupied by all traffic participants during their motion is less than or equal to the number of lanes on the selected road, the road point generation process begins. Subsequently, after the traffic participants adjust their driving direction and initial lane position to match the selected road, the constraint solver generates waypoints based on the driving action constraints.

[0011] To make the participants perform the corresponding actions and drive into the collision site, a set of trajectory constraints C is defined for each action from five aspects. Then, the method automatically generates a trajectory for each participant using a constraint solver. The trajectory is an ordered sequence of waypoints, i.e., positions and velocities that the traffic participants must follow. To formalize the representation of waypoints and design trajectory constraints, we introduce b th The i th th waypoint can be represented as where x, y, v are the x-coordinate, y-coordinate, and velocity, respectively. In addition, we introduce pos = (x, y) to represent the position of the waypoint in the plan view. In addition, the traffic participants must arrive at the collision location simultaneously when performing the action. To do this, the method introduces a collision area CA, which is automatically calculated by the method.

[0012] Group 1: Initial position and destination constraints

[0013] The first group has four constraints. Equation 1 limits the driving direction from to to be the same as the direction of lane l m . The calculation of fd(w i , w j , l) is defined as

[0014] (w j .x-w i .x)(l ex .x-l en .x)>0∧(w j .y-w i .y)(l ex .y-l en .y)>0.

[0015] where l en is the entry point of the lane, and l exThese are the lane exit points, and both points are known after parsing the map. When f d When true, it ensures that in the direction of lane 1, waypoint w j Leading w i Equation 2 restricts the starting and ending road locations for the FollowLane behavior. and In the same lane m Roadside R i Formulas 3 and 4 define the starting and ending road positions for left turns and passing maneuvers: when the driving action avoids a collision, and and Formula 3 applies when on different road lanes; Formula 4 applies when the action results in a collision, destination location. In CA.

[0016]

[0017] Group 2: Positional restrictions for performing actions

[0018] In this group, Equation 5 restricts the position of the waypoint when performing a left turn, ensuring that the current waypoint is to the right of the line connecting the two adjacent waypoints. This represents the direction vector composed of the i-th waypoint position and the (i+1)-th waypoint position. Let k1 represent the right normal vector formed by the positions of waypoints i-th and (i+2)-th. Equations 6 and 7 impose constraints on the positional relationships between roads accessed by performing left-turn and vehicle crossing actions, respectively. k1 and k2 represent road l m and l n The slope.

[0019]

[0020] Group 3: Constraints on Waypoint Speed

[0021] Group 3 imposes speed restrictions on all intermediate waypoints. Equation 8, in particular, restricts... arrive The speed at waypoints is limited by Equation 9. DX(i,j) and DY(i,j) calculate the distances along the X and Y axes from waypoint ith to waypoint jth, respectively. Where vc.x should be the velocity of the waypoint on the X-axis, and vc.y should be the velocity of the waypoint on the Y-axis. spd i The speed is v i .spd limit It is the speed limit for driving on the road.

[0022]

[0023] Group 4: Trajectory combination constraints for multiple basic driving maneuvers

[0024] When multiple maneuvers are connected, the fourth group of constraints restricts the relationship between waypoints. Equations 10 and 11 represent the trajectory constraints for the attacker and victim multiple maneuvers, respectively. The end position of the current maneuver of a traffic participant should be consistent with the start position of the subsequent driving maneuver. The attacker performs n maneuvers, while the victim performs m maneuvers before the collision.

[0025]

[0026] Group 5: Vehicle collision constraints

[0027] The fifth group of constraints ensures that the conflict information is consistent with the report description. Equation 12 restricts the occurrence of the accident collision to the location described in the collision report, while equation 13 describes the simultaneous arrival of the attacker and victim at the collision location.

[0028]

[0029] 3) Simulation and test generation; the simulator receives the waypoints generated by the trajectory planning module, combines the environmental condition information extracted by the LLM, and performs simulation to reconstruct the collision scene described in detail in the accident report. Then this method automatically extracts system-level test cases from the output of the trajectory planning module, allowing the autonomous vehicle to choose different trajectories from those described in the accident report to avoid collisions. For the simulated scene generated based on the accident report, multiple test scenarios are created by designating each traffic participant as the ego vehicle. The starting point of each autonomous vehicle is set to the starting point of the generated waypoints, while the NPC vehicles follow the waypoints provided by the trajectory planning module. Collision scenarios involving autonomous vehicles are recorded during these runs. To minimize false positives, collision scenarios where the speed of the autonomous vehicle is not close to zero are calculated.

[0030] Further, the specific steps of step 1) above are as follows:

[0031] Step 1)-1: Start state;

[0032] Step 1)-2: Input accident report text;

[0033] Step 1)-3: Large language model prompt pattern design;

[0034] Step 1)-4: Large language model information extraction;

[0035] Step 1)-5: Output extracted key information (environment, road, dynamic object);

[0036] Step 1)-6: End state.

[0037] Further, the specific steps of the above step 2) are as follows:

[0038] Step 2)-1: initial state;

[0039] Step 2)-2: input the extracted road information, the extracted dynamic object information, the to-be-tested simulation map Map, and the designed accident constraint C;

[0040] Step 2)-3: the analysis tool analyzes the to-be-tested road information in the Map;

[0041] Step 2)-4: determine whether the extracted road type and the selected analyzed road type match, if yes, execute the next step, otherwise execute step 2)-3;

[0042] Step 2)-5: determine whether the number of lanes occupied by the extracted information is less than or equal to the total number of lanes of the analyzed road, if yes, execute the next step, otherwise execute step 2)-3;

[0043] Step 2)-6: adjust the driving direction and the initial lane position;

[0044] Step 2)-7: perform constraint solving to generate the waypoints;

[0045] Step 2)-8: output the accident trajectory;

[0046] Step 2)-9: end state.

[0047] Further, the specific steps of the above step 3) are as follows:

[0048] Step 3)-1: initial state;

[0049] Step 3)-2: input the extracted environment information and all the accident participant waypoints;

[0050] Step 3)-3: perform accident simulation using the simulator;

[0051] Step 3)-4: select the start point and the end point of the to-be-tested autonomous vehicle in the trajectory in the accident;

[0052] Step 3)-5: generate a simulation test case;

[0053] Step 3)-6: test the autonomous driving system;

[0054] Step 3)-7: output the collision test result;

[0055] Step 3)-8: end state. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1A flowchart of a method for constructing an automatic driving generalization test scene based on an accident report according to an embodiment of the present application.

[0057] Figure 2 A flowchart of information extraction. Figure 1

[0058] Figure 3 A flowchart of trajectory planning. Figure 1

[0059] A flowchart of simulation and test generation. Figure 4 Figure 1 DETAILED DESCRIPTION

[0060] In order to better understand the technical content of the present application, specific embodiments are described below with reference to the accompanying drawings.

[0061] Figure 1 A flowchart of a method for constructing an automatic driving generalization test scene based on an accident report according to an embodiment of the present application.

[0062] A method for constructing an automatic driving generalization test scene based on an accident report, characterized in that it comprises the following steps.

[0063] S1 information extraction, using a large language model to extract accident report text information, and designing the same language prompt mode to extract accurate accident information.

[0064] S2 trajectory planning, planning a trajectory consistent with the accident report information, inputting the extracted information, and outputting the waypoints of all traffic participants by solving the accident constraints.

[0065] S3 simulation and test generation, using the waypoints of the traffic participants to reproduce the simulation accident scene, and converting it into a test case to test the automatic driving system.

[0066] Figure 2 A flowchart of information extraction, accident text information is parsed and extracted. The specific steps are as follows:

[0067] Step 1: initial state; Step 2: input accident report text; Step 3: large language model prompt mode design; Step 4: large language model information extraction; Step 5: output extracted key information (environment, road, dynamic object); Step 6: end state.

[0068] Figure 3 A flowchart of trajectory planning. The analyzed simulation map is matched with the analyzed road information and accident road information, and the trajectories of the traffic participants are solved by formulating 5 constraints. The specific steps are as follows:

[0069] ​​​Step 1: initial state; Step 2: input extracted road information, extracted dynamic object information, to-be-tested simulation map Map, designed accident constraint C; Step 3: the analysis tool analyzes the to-be-tested road information in Map; Step 4: whether the extracted road type and the selected parsed road type match, if yes, the next step is executed, otherwise, step 3 is executed; Step 5: the number of occupied lanes of the extracted information is less than or equal to the total number of parsed road lanes, if yes, the next step is executed, otherwise, step 3 is executed; Step 6: adjust the driving direction and the initial lane position; Step 7: generate a route point by constraint solving; Step 8: output the accident trajectory; Step 9: end state.

[0070] Figure 4 The flowchart for simulation and testing is generated. According to the generated route point information, the traffic accident scene is reproduced, and the accident scene is converted into a simulation test scene to test the automatic driving system. The specific steps are as follows:

[0071] Step 1: initial state; Step 2: input extracted environment information and all accident participant route points; Step 3: use the simulator to simulate the accident; Step 4: select the start point and the end point of the to-be-tested automatic driving vehicle in the trajectory in the accident; Step 5: generate a simulation test case; Step 6: test the automatic driving system; Step 7: output the collision test result; Step 8: end state.

[0072] In summary, the present application introduces a tool for automatically reconstructing an accident scene from an accident report and testing an ADS. The method uses LLM to extract detailed accident information, and by designing a special language mode for extraction prompts, the text understanding and parsing ability of LLM is significantly improved. In order to enhance the generalization ability of the method in the reconstruction of the accident scene, the method generates trajectories related to the accident by solving a set of predefined trajectory specifications. The method can successfully reproduce the accident report on different map structures, and the reconstructed simulation scene can identify various types of safety violations based on industrial-level ADS. The method plays a key role in ensuring the quality and reliability of the ADS.

Claims

1. An automatic driving generalization test scenario construction method based on accident reports, characterized by, According to the layer-based scene definition, the accident scene information is preliminarily abstracted into three layers: road network and traffic guiding objects, environmental conditions and dynamic objects; then the large language model (LLM) is used to systematically extract information about the influencing factors of the accident from the accident report, and the information is organized layer by layer; subsequently, constraints are established for the pre-accident driving behavior of the traffic participants, and a constraint solver is used to generate trajectories that meet the specified constraints in the accident report; then, the extracted environmental and road information and the generated trajectories are input into a driving simulator to reconstruct the car accident scene; finally, in the evaluation stage, the generated simulation scene is converted into test cases containing a test oracle, and these test cases are input into an autonomous driving system (ADS); this method checks whether the tested ADS successfully reaches the final intended position without encountering any crashes; the method comprises the following steps: 1) information extraction; in order to abstractly represent the accident information and organize it into a semantic structure that the LLM can understand, the extracted information is presented in a hierarchical representation, divided into three layers: environment, road and dynamic objects; the environment layer includes weather and lighting conditions; weather conditions (Weather) include rain, fog, snow and other factors; lighting conditions (Lighting) are related to the lighting conditions on the road, which are usually brighter during the day and darker at night; in addition, streetlights can enhance lighting; the road layer describes the road network and traffic signs for road guidance; the road represents the geographical background of the collision, including the type of road where the accident occurred (CollisionLocation) and the number of lanes on the relevant road (LaneNum); in addition, the method extracts information about the speed limit (SpeedLimit) on the road to reconstruct the speed constraints applicable to the accident location; the dynamic object layer contains information about the colliding parties and the victims, as well as the moving actions that led to the collision; the method extracts information about the number of traffic participants involved in the collision (ParticipantsNumber) and identifies the type of collision; the collision type information (CrashType) specifies the angle at which the traffic participants collided, including three types of collisions: rear-end collision, head-on collision and front-back collision; for each vehicle involved in the accident, the method extracts the state of each vehicle, including the initial running lane of the vehicle (RunningLanes), the initial driving direction of the vehicle (DrivingDirections) and the behavior of the vehicle before the collision (DrivingActions); specifically, the vehicle behavior describes the normal and abnormal behavior of the vehicle; routine driving actions include U-turn, stopping, entering the road, vehicle crossing, left turn, right turn, following the lane and changing lanes; abnormal driving behavior includes driving off the road and driving in reverse; the method uses prompt patterns to extract accident information layer by layer; for each language pattern, the meaning of each attribute is first explained to help the LLM understand the extracted information; In addition, the model also includes heuristic rules to guide the LLM to produce accurate results; Finally, the use of few-shot learning to ensure that the output of the LLM meets the expected standards; 2) Trajectory planning; This method uses the extracted driving actions and road information to simulate collisions and calculate the trajectories of simulated traffic participants; The road information R extracted in step 1), the dynamic object information D extracted in step 1), the given map MAP, and the defined driving action constraints C are input; First, parse the given map into a set of candidate roads, and note the road type of each road; Then iterate through the candidate road set, and select lanes of different lengths and widths in turn; If the current road type matches the road type where the accident occurred, and the maximum number of lanes occupied by all traffic participants during their movement is less than or equal to the number of lanes on the selected road, then start the waypoint generation process; Subsequently, after the traffic participants adjust their driving direction and initial lane position to match the selected road, the constraint solver generates the waypoints based on the driving action constraints; To make the participants perform the corresponding actions and drive into the crash scene, a set of trajectory constraints C is defined for each action from five aspects; then, the method automatically generates trajectories for each participant using a constraint solver; the trajectory is an ordered sequence of waypoints, i.e., positions and velocities that the road user must follow; to formalize the waypoints and design the trajectory constraints, b th i th The waypoint is represented as where x, y, v are the x-coordinate, y-coordinate, and velocity, respectively; in addition, we introduce pos = (x, y) to represent the position of the waypoint in the plan view; in addition, the road user must arrive at the crash location simultaneously when performing the action; to do this, the method introduces a crash area CA, which is automatically calculated by the method; Group 1: Initial position and destination constraints The first set has four constraints; equation 1 restricts the direction of travel from to to be the same as the direction of lane l m ; the calculation fd(w i , w j , l) is defined as (w j .x-w i .x)(l ex .x-l en .x) > 0 A (w j .y-w i .y)(l ex .y-l en .y) > 0; where l en is the entry point of the lane, l ex is the exit point of the lane, both points are known after parsing the map; when f d is true, it guarantees that the waypoint w j is ahead of w i in the direction of the lane 1 ; equation 2 limits the road position where the Follow Lane behavior starts and ends; and in the same lane l m road side R i ; formulas 3 and 4 define the start and end road position for the left turn and the pass actions: formula 3 is applied when the driving action avoids a collision, and and formula 4 is adopted when the action leads to a collision, the destination position is in CA; Group 2: Position constraints for performing actions In this group, the formula 5 limits the position of the waypoint when performing the left turn action, ensuring that the current waypoint is located on the right side of the line connecting the adjacent two waypoints; represents the direction vector composed of the i-th waypoint position and the (i+1)-th waypoint position; vector represents the right normal vector formed by the positions of the i-th and (i+2)-th waypoints; the equations 6 and 7 impose restrictions on the positional relationship between the road accessed by performing the left turn behavior and the vehicle crossing behavior, respectively; k1 and k2 represent the slopes of the roads l m and l n ; Group 3: Constraints on waypoint speed Group 3 limits the speed of all intermediate waypoints; wherein, Equation 8 limits the speed of the waypoint to Equation 9 limits the speed limit while driving; DX(i,j) and DY(i,j) calculate the distance of the vehicle along the X-axis and Y-axis from the ith waypoint to the jth waypoint, respectively; wherein, vc.x should be the speed of the waypoint along the X-axis, vc.y is the speed of the waypoint along the Y-axis; spd i of the vehicle is υ i spd limit is the speed limit of the driving road; Group 4: Trajectory combination constraints for multiple basic driving actions When multiple operations are connected, the fourth group constrains the relationship between waypoints; Equations 10 and 11 represent the trajectory constraints for the attacker and victim multiple actions, respectively; The end position of the current action of the traffic participant should match the start position of the subsequent driving action; The attacker performs n actions, while the victim performs m actions before the collision; Group 5: Vehicle collision constraints The fifth group of constraints ensures that the conflict information matches the description in the report; Equation 12 limits the occurrence of the accident collision to the position described in the collision report, while equation 13 describes the simultaneous arrival of the attacker and victim at the collision location; 3) Simulation and test generation; The simulator receives the waypoints generated by the trajectory planning module and combines them with the environmental condition information extracted by the LLM to perform simulations and reconstruct the collision scene described in detail in the accident report; Then this method automatically extracts system-level test cases from the output of the trajectory planning module, allowing the autonomous vehicle to choose a different trajectory than the one described in the accident report to avoid a collision; For the simulated scene generated based on the accident report, multiple test scenarios are created by designating each traffic participant as the ego vehicle; The starting point of each autonomous vehicle is set to the starting point of the generated waypoints, while the NPC vehicles follow the waypoints provided by the trajectory planning module; During these runs, collision scenarios involving autonomous vehicles are recorded; To minimize false positives, collision scenarios where the speed of the autonomous vehicle is not close to zero are calculated. 2.The accident report based automated driving generalization test scenario construction method of claim 1, wherein, In step 1), information extraction is performed to extract valuable information about the road environment and vehicle trajectories contained in the accident report using LLM; during the information extraction process, the narrative of the car accident is incrementally parsed, and information about the weather, lighting, road, and vehicle is accumulated into a data structure to form a summary of the car accident; a suitable prompt is designed to accurately describe the content to be queried or requested to improve the extraction accuracy and effectiveness of the LLM. 3.The accident report based automated driving generalization test scenario construction method of claim 1, wherein, In step 2), trajectory planning is performed to simulate the collision and calculate the trajectory of the simulated traffic participants using the extracted driving actions and road information; the constraint solver analyzes the constraints according to the set driving actions to generate a sequence of waypoints for the trajectory. 4.The method of claim 1, wherein, In step 3), simulation and test generation are performed to reproduce the collision scenario in the accident report using the waypoints from the trajectory planning module and the environmental information provided by the LLM; the method automatically creates system-level test cases by setting one of the traffic participants as an autonomous vehicle, setting the starting point and endpoint according to the trajectory planning, and solving the waypoint sequence to generate the test scene and record the collision situation.

Images