Unmanned driving extreme case generation and verification method based on simulation engine

Abstract

The invention discloses an unmanned driving extreme case generation and verification method based on a simulation engine. The method comprises the following steps: S1, constructing a high-fidelity virtual simulation environment; s2, initializing a parameterized model generated by an extreme driving case; s3, generating an initial extreme driving case population based on a genetic algorithm; s4, collecting test result data of the unmanned driving system; s5, quantifying the test effect of each extreme driving case; s6, based on a fitness evaluation result, performing selection, crossover and mutation operation through a genetic algorithm; and S7, the extreme driving case population generated by evolution is subjected to unmanned driving system testing again in the simulation environment, S4 to S6 are executed repeatedly until a preset termination condition is reached, and the termination condition comprises that the scene diversity reaches a target value or the fitness is converged. According to the invention, the generation efficiency and coverage range of the extreme driving scene are improved.

Description

Technical Field

[0001] The present invention relates to the field of driverless technology, and particularly to a method for generating and verifying extreme cases of driverless based on a simulation engine. Background Art

[0002] With the rapid development of artificial intelligence and driverless technology, driverless systems are gradually becoming an important part of intelligent transportation. The core of driverless technology lies in the system's perception of the environment, path planning, and behavior decision-making. The reliability and safety of these modules are directly related to the actual performance of driverless vehicles in complex scenarios. Therefore, the verification of system performance under extreme driving conditions is particularly important.

[0003] Currently, for the testing of driverless systems, traditional methods mainly rely on manually designed scenarios or road tests based on the real world. Manually designed scenarios require R & D personnel to define various test conditions according to experience, such as complex road conditions, sudden obstacles, and bad weather. The disadvantage of this method is that scenario design usually fails to cover all extreme situations that may be encountered in actual driving, especially the low-probability but high-risk "long-tail" problems.

[0004] In recent years, some research has started to combine simulation technology with driverless testing. By using a simulation engine to construct a virtual driving environment and generating test scenarios through simple rule settings, however, there are limitations in the diversity and complexity of the generated scenarios: on the one hand, existing scenario generation technologies usually rely on fixed rules or manual configuration and cannot dynamically generate complex and diverse extreme driving scenarios; on the other hand, the verification metrics of simulation tests mostly focus on a single dimension, such as the accuracy of the perception module, and lack multi-dimensional verification of the entire perception, decision-making, and execution link.

[0005] In summary, the existing technologies have significant deficiencies in the coverage of test scenario generation, the efficiency of the test process, and the comprehensive verification ability of system performance, and cannot meet the high standards of reliability and safety requirements of driverless systems under extreme driving conditions. Therefore, there is an urgent need for a test method that can efficiently generate diverse extreme driving scenarios and comprehensively verify the performance of driverless systems to solve the above technical problems. Summary of the Invention

[0006] The purpose of the present invention is to propose a method for generating and verifying extreme cases of driverless based on a simulation engine, which can effectively solve the problems of single test scenarios, limited verification dimensions, and low test efficiency in the existing technologies. It can not only improve the generation efficiency and coverage of extreme driving scenarios, but also realize a multi-dimensional and full-link system verification process.

[0007] The technical solution adopted by the present invention to achieve the above object is: a method for generating and verifying extreme cases of driverless based on a simulation engine, including the following steps:

[0008] S1. Construct a high-fidelity virtual simulation environment;

[0009] S2. Initialize the parametric model for generating extreme driving cases. The parametric model is based on preset scenario rules, and the scenario rules are designed according to traffic regulations, actual driving behaviors, and known dangerous patterns;

[0010] S3. Generate an initial population of extreme driving cases based on the genetic algorithm. The initial population contains multiple extreme driving cases. The parameters of the extreme driving cases are generated by random initialization or rule setting, and each extreme driving case is described by a set of scenario parameters;

[0011] S4. Run the initial population of extreme driving cases in the simulation environment, test the driverless system for each extreme driving case, and collect the test result data of the driverless system;

[0012] S5. Evaluate the fitness of the population of extreme driving cases according to the test result data. The fitness evaluation is based on a multi-index system to quantify the test effect of each extreme driving case;

[0013] S6. Based on the fitness evaluation results, perform selection, crossover, and mutation operations through the genetic algorithm. Select the extreme driving cases with fitness higher than the threshold to enter the next generation. The crossover operation is used to generate new combinations of scenario parameters, and the mutation operation is used to introduce random perturbations to increase scenario diversity;

[0014] S7. Test the evolved population of extreme driving cases in the simulation environment again for the driverless system, and repeat steps S4 to S6 until the preset termination conditions are met. The termination conditions include that the scenario diversity reaches the target value or the fitness converges.

[0015] Optionally, S1 includes the following specific steps:

[0016] S11. Construct a road structure model, and the road structure model includes straight roads, curved roads, intersections, and slope sections;

[0017] S12. Construct a dynamic traffic scenario model, and the dynamic traffic scenario model is based on the kinematic characteristics and interaction rules of virtual vehicles. The interaction rules include following behavior, lane-changing behavior, and obstacle avoidance behavior;

[0018] S13. Construct a physical interaction model, and the physical interaction model is defined by simulating the dynamic interaction relationship between the vehicle and the environment;

[0019] S14. Set adjustable environmental parameters, including the environmental parameters, traffic flow density, and obstacle position and type:

[0020] The environmental parameters include weather conditions, defined as light intensity, rainfall, snowfall, and wind speed;

[0021] The traffic flow density is defined as the flow density of vehicles per unit time;

[0022] The obstacle position and type, where the obstacle position is represented by coordinates (x o , y o ), and the obstacle types include static obstacles and dynamic obstacles;

[0023] S15. Combine the road structure model, dynamic traffic scene model, physical interaction model, and adjustable environmental parameters to generate a high-fidelity virtual simulation environment.

[0024] Optionally, the S2 includes the following specific steps:

[0025] S21. Initialize the initial state of the scene, where the initial state of the scene is based on parametric descriptions:

[0026] Road geometric characteristic parameters (w d , r c , θ s , l d ), where w d represents the road width, r c represents the radius of curvature, θ s represents the slope angle, and l d represents the road length;

[0027] The position and state of traffic participants, defined as the initial position coordinates of virtual vehicles speed acceleration and direction angle

[0028] Environmental parameters, including weather conditions where represents the initial light intensity, represents the initial rainfall, represents the initial snowfall, represents the initial wind speed;

[0029] S22. Define dynamic change rules, including vehicle kinematic dynamics:

[0030]

[0031] where, and Denote the position of the vehicle at time t+1, and Δt denote the time step;

[0032] The environment changes dynamically, including the light intensity I w , the rainfall P r , the snowfall P s and the wind speed v w The change rules over time are as follows:

[0033]

[0034] Among them, α and β represent the rates of dynamic changes;

[0035] I w (t) = I base *a weather

[0036] Among them, I base is a local reference light intensity, and a weather is the weather attenuation factor;

[0037] v w = v0 + Δv

[0038] Among them, v0 is the basic wind speed, and Δv is the wind speed amplitude.

[0039] The behavior rules of obstacles are defined as that static obstacles maintain their positions unchanged, and dynamic obstacles move along specific paths:

[0040]

[0041] Among them, and denote the position of the dynamic obstacle at time t+1, v o denotes the obstacle speed, and φ o denotes the obstacle direction angle;

[0042] S23. Set the termination conditions:

[0043] The scene completion condition is that all traffic participants reach their target positions or the simulation running time reaches the preset upper limit;

[0044] The system status condition is that the driverless system has a collision, path planning fails, or there is a timeout response;

[0045] The environmental stability condition is that the dynamic parameters reach a stable state or exceed the set threshold range.

[0046] S24. Generate a parametric model based on the initial scene state, dynamic change rules, and termination conditions.

[0047] Optionally, the S3 includes the following specific steps:

[0048] S31. Define the initial population size N and population structure. The initial population consists of N extreme driving cases. The parameters of each extreme driving case include road geometric characteristic parameters (w d , r c , θ s , l d ), traffic participant status parameters environmental parameters and obstacle parameters

[0049] S32. Generate the initial population through random initialization. The random initialization is based on the set parameter ranges, and assigns random parameter values to each extreme driving case:

[0050] Road width w d ∈ [3, 10] m, radius of curvature r c ∈ [10, 200] m, slope angle θ s ∈ [-15°, 15°], road length l d ∈ [50, 500] m;

[0051] Initial position coordinates of traffic participants m, initial speed m / s, initial acceleration m / s2, initial direction angle rad;

[0052] Illumination intensity lm, rainfall mm / h, snowfall mm / h, wind speed m / s;

[0053] Initial position coordinates of obstacles m, speed v o ∈ [0, 10] m / s, direction angle φ o ∈ [0, 2π] rad;

[0054] S33. Generate specific extreme driving cases based on rule settings;

[0055] S34. Conduct a legality check on the initial population, eliminate the extreme driving cases that do not meet the constraint conditions, and output the initial population of extreme driving cases that pass the legality check.

[0056] Optionally, the generation of specific extreme driving cases based on rule settings includes the following categories:

[0057] High-risk road condition rules:

[0058] Curved road scenario: Set the radius of curvature rc ∈[10, 50] m to simulate a sharp turn scenario;

[0059] Sloped road section scenario: Set the slope angle θ s ∈[-15°, -10°] ∪ [10°, 15°] to simulate the dynamic impact of uphill and downhill on driverless vehicles;

[0060] Rainfall scenario: Set the rainfall mm / h² to simulate sudden heavy rain;

[0061] Wind speed scenario: Set the wind speed m / s² to simulate the impact of crosswind or tailwind on vehicle stability;

[0062] Complex traffic flow scenario rules

[0063] Multiple vehicle rear-end collision scenario: Set the initial position spacing d of virtual vehicles gap ∈[0, 5] m, speed difference v diff ∈[5, 15] m / s to simulate a chain rear-end collision accident on the highway;

[0064] Complex intersection interaction scenario: Set the initial coordinates of multiple vehicles converging at an intersection Speed m / s, direction angle To simulate the interaction dynamics in case of signal failure;

[0065] Extreme weather condition rules:

[0066] Blizzard scenario: Set the snowfall mm / h² to simulate the condition of reduced road surface sliding coefficient;

[0067] Low visibility scenario: Set the light intensity lumens to simulate the sensor perception ability in haze weather;

[0068] Obstacle sudden behavior rules:

[0069] Dynamic obstacle: Set the initial position of the obstacle Speed v o ∈[2, 8] m / s, direction angle φ o ∈[0, π] to simulate the scenario of pedestrians crossing the road;

[0070] Sudden obstacle: Dynamically generate an obstacle, and the position change rule is:

[0071]

[0072] Among them, the speed v o ∈[5, 10] m / s, and the direction angle changes with time Denote the disturbance angle;

[0073] Rules for vehicle dynamics failure scenarios:

[0074] Simulate the scenario of weakened tire grip: Set the friction coefficient μ ∈ [0.2, 0.4] to simulate vehicle skidding caused by a wet road surface;

[0075] Simulate emergency braking of the vehicle: Set the vehicle acceleration a v ∈ [-10, -5] m / s² to evaluate the emergency response ability of the driverless system;

[0076] Rules for special scenario combinations:

[0077] Combine dynamic traffic flow and extreme weather scenarios to simulate multi-vehicle rear-end collisions in a heavy rain scenario;

[0078] Combine the uphill section and dynamic obstacle scenarios to study the impact on the vehicle's decision-making path when an obstacle suddenly moves to the middle of the uphill.

[0079] Optionally, the S4 includes the following specific steps:

[0080] S41. Load the initial population of extreme driving cases in a high-fidelity simulation environment. The parameters of each extreme driving case include road geometric characteristic parameters, traffic participant state parameters, environmental parameters, and obstacle parameters;

[0081] S42. Connect the driverless system to the simulation environment and run each extreme driving case;

[0082] S43. Collect the recognition accuracy η of the perception module during the operation of the driverless system p :

[0083]

[0084] The decision-making rationality of the path planning module, measured by the path error ∈ p :

[0085]

[0086] The response time τ of the execution module e , representing the time delay from the input of the perception module to the vehicle's execution of an action;

[0087] Vehicle state changes, including position, speed, acceleration, and steering angle.

[0088] Optionally, the S5 includes the following specific steps:

[0089] S51. Establish a multi-index evaluation system and calculate the following key evaluation indicators for each extreme driving case:

[0090] Calculate the successful risk avoidance rate η by comprehensively considering the collision situation and environmental dynamic factors s :

[0091]

[0092] where 1(d t ≤0) indicates whether a collision event occurs at time t, and d t is the minimum safety distance between the driverless vehicle and obstacles or other vehicles respectively represent the light intensity, rainfall, snowfall, and wind speed at time t, α1 is the environmental risk coefficient, and γ1 is the environmental disturbance weight

[0093] The number of collisions C sum , used to measure the cumulative situation of collision events in extreme driving cases

[0094]

[0095] The path planning error ∈ C , representing the average deviation between the actual driving trajectory of the driverless vehicle and the planned path. Refer to the form of the path error and introduce the adaptive weight γ2 to consider the road geometric characteristic parameters (w d , r c , θ s , l d ):

[0096]

[0097] where γ2 represents the modulation coefficient for the slope angle θ s and the radius of curvature r c ;

[0098] The vehicle dynamic response robustness η r , representing the ability of the vehicle to maintain stable driving under adverse environmental parameters, is defined by comprehensively considering sensor fluctuations and actuator module delays as

[0099]

[0100] where represents the response time of the actuator module at time t represents the acceleration mutation amplitude of the vehicle at time t, α2 is the time delay weight of the dynamic response, and γ3 is the acceleration disturbance weight

[0101] S52. Based on the multi-index evaluation system, calculate the comprehensive fitness value F of each extreme driving case i : F i = w s ·

[0102] η s -w c ·C sum -w p ·∈ C +w r ·η r ;

[0103] Among them, w s , w c , w p , w r are the weighting coefficients of each index, and w s +w c +w p +w r = 1. The weighting coefficients can be configured according to actual test requirements in different extreme scenarios;

[0104] S53. Sort the extreme driving case population according to the comprehensive fitness value F i , and use the sorting function Ψ(·) to assign index to all extreme driving cases:

[0105]

[0106] Among them, Sort means sorting the fitness values of all extreme driving cases from high to low, and Ψ(F i ) is the sorting rank corresponding to the fitness value, and N c represents the total number of extreme driving cases participating in the evaluation.

[0107] Optionally, the S6 includes the following specific steps:

[0108] S61. Determine the fitness threshold F th , and perform a selection operation based on the comprehensive fitness value F i , and select all extreme driving cases that satisfy F i ≥F th to form a candidate set Ω1;

[0109] S62. Randomly pair the extreme driving cases in the candidate set Ω1 in pairs and perform a crossover operation to generate new combinations of scenario parameters;

[0110] S63. Perform a mutation operation on the offspring scenario parameters generated after crossover, and introduce random perturbations to increase scenario diversity;

[0111] S64. Combine the offspring extreme driving cases generated after the crossover and mutation operations with the parent extreme driving cases to form a new population, which is used as the input of the next generation of extreme driving cases.

[0112] The beneficial effects of the present invention are:

[0113] (1) The present invention realizes the dynamic optimization of extreme driving cases through a genetic algorithm, generates diverse extreme driving scenarios by combining fitness evaluation and evolution paths, can dynamically generate complex scenario parameter combinations through intelligent crossover and mutation operations, and at the same time achieves a more efficient coverage rate for low-probability and high-risk scenarios in the "long tail" problem. By introducing an adaptive perturbation mechanism, the simulation environment can generate complex conditions that are closer to actual driving situations, improving the flexibility and richness of scenario generation, and effectively making up for the deficiencies of existing technologies in scenario coverage.

[0114] (2) The present invention designs a multi-dimensional fitness evaluation index system, including the successful risk avoidance rate, path planning error, and vehicle dynamic response robustness, to comprehensively analyze the full-link performance of the unmanned driving system. The index system can simultaneously evaluate the collaborative effects of the perception, decision-making, and execution modules, provide more comprehensive performance evaluation data, adjust the influence of each evaluation index through a dynamic weight mechanism, optimize the rationality of fitness calculation, quickly locate and improve the performance bottleneck of the unmanned driving system, and effectively shorten the test and verification cycle.

[0115] (3) The present invention supports modular design. By combining a parameterized model and a simulation engine, it can achieve controllable generation and accurate reproduction of extreme scenarios. Through parameterized description and dynamic regulation, it can reproduce complex driving scenarios in the simulation environment at low cost and high efficiency, providing accurate data support for the algorithm debugging and performance optimization of the unmanned driving system. In addition, the introduction of an automated test framework greatly reduces manual intervention, ensuring the repeatability of test results and the objectivity of verification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0116] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, but do not constitute a limitation to the present invention. In the drawings:

[0117] Figure 1 is a flowchart of a method for generating and verifying extreme cases of unmanned driving based on a simulation engine proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0118] The present invention will now be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0119] Refer to Figure 1 , a method for generating and verifying extreme cases of unmanned driving based on a simulation engine, includes the following steps:

[0120] S1. Construct a high-fidelity virtual simulation environment;

[0121] S2. Initialize the parameterized model for generating extreme driving cases. The parameterized model is based on preset scenario rules, and the scenario rules are designed according to traffic regulations, actual driving behaviors, and known dangerous patterns.

[0122] S3. Generate an initial population of extreme driving cases based on the genetic algorithm. The initial population contains multiple extreme driving cases. The parameters of the extreme driving cases are generated by random initialization or rule setting, and each extreme driving case is described by a set of scenario parameters.

[0123] S4. Run the initial population of extreme driving cases in the simulation environment, conduct unmanned driving system tests on each extreme driving case, and collect the test result data of the unmanned driving system.

[0124] S5. Evaluate the fitness of the extreme driving case population according to the test result data. The fitness evaluation is based on a multi-index system to quantify the test effect of each extreme driving case.

[0125] S6. Based on the fitness evaluation results, perform selection, crossover, and mutation operations through the genetic algorithm. Select the extreme driving cases with fitness higher than the threshold to enter the next generation. The crossover operation is used to generate new combinations of scenario parameters, and the mutation operation is used to introduce random perturbations to increase scenario diversity.

[0126] S7. Conduct unmanned driving system tests on the evolved population of extreme driving cases in the simulation environment again, and repeat S4 to S6 until the preset termination conditions are reached. The termination conditions include the scenario diversity reaching the target value or the fitness converging.

[0127] In this embodiment, S1 includes the following specific steps:

[0128] S11. Construct a road structure model, which includes straight roads, curved roads, intersections, and slope sections.

[0129] S12. Construct a dynamic traffic scenario model, which is based on the kinematic characteristics and interaction rules of virtual vehicles. The interaction rules include following behavior, lane-changing behavior, and obstacle avoidance behavior.

[0130] S13. Construct a physical interaction model, which is defined by simulating the dynamic interaction relationship between the vehicle and the environment.

[0131] S14. Set adjustable environment parameters, including environmental parameters, traffic flow density, and the position and type of obstacles:

[0132] The environmental parameters include weather conditions, which are defined as light intensity, rainfall, snowfall, and wind speed.

[0133] The traffic flow density is defined as the flow density of vehicles per unit time.

[0134] Obstacle position and type. The obstacle position is represented by coordinates (x o , y o ), and the obstacle types include static obstacles and dynamic obstacles;

[0135] S15. Generate a high-fidelity virtual simulation environment by combining a road structure model, a dynamic traffic scene model, a physical interaction model, and adjustable environmental parameters.

[0136] In this embodiment, S2 includes the following specific steps:

[0137] S21. Initialize the initial state of the scene, and the initial state of the scene is based on parametric descriptions:

[0138] Road geometric characteristic parameters (w d , r c , θ s , l d ), where w d represents the road width, r c represents the radius of curvature, θ s represents the slope angle, and l d represents the road length;

[0139] Traffic participant position and state, defined as the initial position coordinates of a virtual vehicle speed acceleration and direction angle

[0140] Environmental parameters, including weather conditions where represents the initial light intensity, represents the initial rainfall, represents the initial snowfall, represents the initial wind speed;

[0141] S22. Define dynamic change rules, including vehicle kinematic dynamics:

[0142]

[0143] where, and represent the position of the vehicle at time t + 1, and Δt represents the time step;

[0144] Environmental dynamic changes, including light intensity I w , rainfall P r , snowfall P s and wind speed v w rules of change over time:

[0145]

[0146] Among them, α and β represent the rates of dynamic change;

[0147] I w (t) = I base *a weather

[0148] Among them, I base is a local reference light intensity, and a weather is the weather attenuation factor;

[0149] v w = v0 + Δv

[0150] Among them, v0 is the basic wind speed, and Δv is the wind speed amplitude.

[0151] The obstacle behavior rules are defined as that static obstacles maintain their positions unchanged, and dynamic obstacles move along specific paths:

[0152]

[0153] Among them, and represent the position of the dynamic obstacle at time t + 1, v o represents the obstacle speed, and φ o represents the obstacle direction angle;

[0154] S23. Set the termination conditions:

[0155] Scenario completion condition: all traffic participants reach their target positions or the simulation running time reaches the preset upper limit;

[0156] System status condition: the driverless system has a collision, path planning fails, or there is a timeout response;

[0157] Environmental stability condition: the dynamic parameters reach a stable state or exceed the set threshold range.

[0158] S24. Generate a parametric model based on the initial state of the scenario, the dynamic change rules, and the termination conditions.

[0159] In this embodiment, S3 includes the following specific steps:

[0160] S31. Define the initial population size N and the population structure. The initial population consists of N extreme driving cases. The parameters of each extreme driving case include road geometric characteristic parameters (w d , r c , θ s , l d ), traffic participant status parameters Environmental parameters and obstacle parameters

[0161] S32. Generate an initial population through random initialization. The random initialization is based on the set parameter ranges, and random parameter values are assigned to each extreme driving case:

[0162] Road width w d ∈ [3, 10] m, radius of curvature r c ∈ [10, 200] m, slope angle θ s ∈ [-15°, 15°], road length l d ∈ [50, 500] m;

[0163] Initial position coordinates of the virtual vehicle m, initial speed m / s, initial acceleration m / s², initial direction angle rad;

[0164] Illumination intensity lumens, rainfall mm / h, snowfall mm / h, wind speed m / s;

[0165] Initial position coordinates of the obstacle m, speed v o ∈ [0, 10] m / s, direction angle φ o ∈ [0, 2π] rad;

[0166] S33. Generate specific extreme driving cases based on rule settings;

[0167] S34. Conduct a legality check on the initial population, eliminate the extreme driving cases that do not meet the constraint conditions, and output the initial population of extreme driving cases that pass the legality check.

[0168] In this embodiment, generating specific extreme driving cases based on rule settings includes the following categories:

[0169] High-risk road condition rules:

[0170] Curved road scenario: Set the radius of curvature r c ∈ [10, 50] m to simulate a sharp turn scenario;

[0171] Sloped road section scenario: Set the slope angle θ s ∈ [-15°, -10°] ∪ [10°, 15°] to simulate the dynamic impact of uphill and downhill on the driverless vehicle;

[0172] Rainfall scenario: Set the rainfall to 2 millimeters per hour to simulate sudden heavy rain;

[0173] Wind speed scenario: Set the wind speed to 2 meters per second to simulate the impact of crosswind or tailwind on vehicle stability;

[0174] Complex traffic flow scenario rules

[0175] Multi-vehicle rear-end collision scenario: Set the initial position spacing d of the virtual vehicles gap ∈ [0, 5] meters, and the speed difference v diff ∈ [5, 15] meters per second to simulate a chain rear-end collision accident on the highway;

[0176] Complex intersection interaction scenario: Set the initial coordinates of multiple vehicles converging at an intersection Speed in meters per second, and the direction angle to simulate the interaction dynamics in the case of traffic signal failure;

[0177] Extreme weather condition rules:

[0178] Blizzard scenario: Set the snowfall to 2 millimeters per hour to simulate the condition of reduced road surface sliding coefficient;

[0179] Low visibility scenario: Set the light intensity in lumens to simulate the sensor perception ability in foggy and hazy weather;

[0180] Obstacle sudden behavior rules:

[0181] Dynamic obstacle: Set the initial position of the obstacle Speed v o ∈ [2, 8] meters per second, and the direction angle φ o ∈ [0, π] to simulate the scenario of pedestrians crossing the road;

[0182] Sudden obstacle: Dynamically generate an obstacle, and the position change rule is:

[0183]

[0184] Among them, the speed v o ∈ [5, 10] meters per second, and the direction angle changes with time indicating the disturbance angle;

[0185] Vehicle dynamics failure scenario rules:

[0186] Simulate the scenario of reduced tire grip: Set the friction coefficient μ ∈ [0.2, 0.4] to simulate vehicle skidding caused by a wet and slippery road surface;

[0187] Simulate emergency braking of the vehicle: Set the vehicle acceleration a v ∈[-10, -5] m / s² to evaluate the emergency response ability of the driverless system;

[0188] Special scenario combination rules:

[0189] Combine dynamic traffic flow and extreme weather scenarios, and simulate multi-vehicle rear-end collisions in heavy rain scenarios;

[0190] Combine ramp sections and dynamic obstacle scenarios, and study the impact on the vehicle's decision-making path when an obstacle suddenly moves to the middle of the ramp.

[0191] In this embodiment, S4 includes the following specific steps:

[0192] S41. Load the initial extreme driving case population in a high-fidelity simulation environment. The parameters of each driving case include road geometric characteristic parameters, traffic participant state parameters, environmental parameters, and obstacle parameters;

[0193] S42. Connect the driverless system to the simulation environment and run each extreme driving case;

[0194] S43. Collect the recognition accuracy η of the perception module during the operation of the driverless system p :

[0195]

[0196] The decision rationality of the path planning module, measured by the path error ∈ p :

[0197]

[0198] The response time τ of the execution module e , representing the time delay from the input of the perception module to the vehicle's execution of an action;

[0199] Vehicle state changes, including position, speed, acceleration, and steering angle.

[0200] In this embodiment, S5 includes the following specific steps:

[0201] S51. Establish a multi-index system and calculate the following key evaluation indicators for each extreme driving case:

[0202] Calculate the successful risk avoidance rate η by comprehensively considering the collision situation and environmental dynamic factors s :

[0203]

[0204] where 1(d t≤0) indicates whether a collision event occurs at time t, d t is the minimum safe distance between the driverless vehicle and obstacles or other vehicles, respectively represent the light intensity, rainfall, snowfall and wind speed at time t, α1 is the environmental risk coefficient, and γ1 is the environmental disturbance weight;

[0205] The number of collisions C sum , used to measure the cumulative situation of collision events in extreme driving cases:

[0206]

[0207] The path planning error ∈ C , which represents the average deviation between the actual driving trajectory of the driverless vehicle and the planned path. Referring to the form of the reference path error and introducing the adaptive weight γ2 to consider the road geometric characteristic parameters (w d , r c , θ s , l d ):

[0208]

[0209] Among them, γ2 represents the modulation coefficient for the slope angle θ s and the radius of curvature r c ;

[0210] The vehicle dynamic response robustness η r , which represents the ability of the vehicle to maintain stable driving under adverse environmental parameters, and is defined by comprehensively considering sensor fluctuations and actuator module delays as:

[0211]

[0212] Among them, represents the response delay of the actuator module at time t, represents the acceleration mutation amplitude of the vehicle at time t, and α2 and γ3 are respectively the time-delay weight and acceleration perturbation weight of the dynamic response;

[0213] S52. Calculate the comprehensive fitness value F of each extreme driving case based on the multi-index system i :

[0214] F i = w s ·η s - w c ·C sum - w p ·∈ C + w r ·η r ;

[0215] Among them, ws , w c , w p , w r are the weighted coefficients for each index, and w s + w c + w p + w r = 1. The weighted coefficients can be configured according to actual test requirements in different extreme scenarios

[0216] S53. According to the comprehensive fitness value F i sort the population of extreme driving cases, and use the sorting function Ψ(·) to assign indexes to all extreme driving cases:

[0217] Ψ(F i ) = Index(Sort(F1, F2, …, F Nc ));

[0218] where Sort represents sorting the fitness values of all extreme driving cases from high to low, Ψ(F i ) is the sorting rank corresponding to the fitness value, and N c represents the total number of extreme driving cases participating in the evaluation

[0219] In this embodiment, S6 includes the following specific steps:

[0220] S61. Determine the fitness threshold F th , and perform a selection operation based on the comprehensive fitness value F i , and select all extreme driving cases that satisfy F i ≥ F th to form a candidate set Ω1;

[0221] S62. Randomly pair the extreme driving cases in the candidate set Ω1 in pairs and perform a crossover operation to generate new combinations of scenario parameters;

[0222] S63. Perform a mutation operation on the offspring scenario parameters generated after crossover, and introduce random perturbations to increase scenario diversity;

[0223] S64. Combine the offspring extreme driving cases generated after crossover and mutation operations with the parent extreme driving cases to form a new population, which is used as the input of the next generation of extreme driving cases

[0224] Example

[0225] It is planned to conduct tests on a high-speed special bridge in the mountainous area in the north of the virtual city. This section contains a complex terrain of continuous sharp curves and long downhill slopes. The minimum curvature radius of the continuous sharp curves is 180 meters, and the slope of the long downhill slope is 8%.

[0226] S1. Construct a high-fidelity virtual simulation environment: By integrating a road structure model (considering geometric characteristics such as a minimum curvature radius of 180 meters and a slope of 8%), a dynamic traffic scenario model (covering multi-lane mixed traffic and following, lane-changing, and obstacle avoidance rules for various vehicles), a physical interaction model (simulating dynamic relationships such as between vehicles and the road surface, gravel and the vehicle body, and the action of strong winds), and adjustable environmental parameters (typhoon rainstorms with an instantaneous rainfall of 90 mm / hour, crosswinds with a gust intensity of level 14, lightning pulse intensity of 50 V / m, etc.), the road characteristics of sharp curves and long downhill slopes and the impact of extreme meteorological disturbances on the unmanned driving system are highly faithfully restored in the bridge area. Through the integration of multi-dimensional simulation elements, a reliable test basis is provided for the generation and iteration of subsequent extreme driving cases.

[0227] S2. Initialize the parametric model for generating extreme driving cases: After completing the construction of the high-fidelity simulation environment, based on the above scenario elements, initialize the parametric model required for extreme driving cases. First, set the test period from 18:30 to 19:30 in the evening, with the external light intensity gradually decreasing; at the same time, due to continuous precipitation, a runoff layer about 3 cm deep is formed on the bridge deck, resulting in a significant reduction in the road surface friction coefficient to 0.25. Assign an initial speed range of 80 to 90 km / h to traffic participants and define dynamic change rules to simulate lightning interference, typhoon wind speed, and rainfall fluctuations over time; if events such as landslides or lateral displacement of freight trains are detected, the road traffic conditions are updated in real time. The scenario termination conditions are set as the vehicle successfully passing through the special bridge or the simulation time reaching 1 hour. If the unmanned driving system experiences collisions, path planning failures, or any environmental parameter violations, the termination mechanism will also be triggered. Thus, a set of reusable parametric scenario descriptions can be obtained for the automatic generation and iteration of subsequent extreme driving cases.

[0228] S3. Generate the initial extreme driving case population based on the genetic algorithm: Further, based on the above parametric model, set the initial population size N = 20 to 30 extreme scenarios, each scenario containing multi-dimensional information such as road geometric characteristics (sharp curve curvature and slope), vehicle state, extreme weather parameters, and obstacle trigger timing. In the random initialization stage, appropriate fluctuation ranges are given to the curvature radius, slope angle, gust intensity, lightning level, etc. to enhance scenario diversity; at the same time, combined with rule settings, emergencies such as gravel sliding or a freight train invading the carriageway by 1.2 meters are simulated in the northern section of the bridge, and a device failure scenario with a 300-millisecond data packet loss in the vision module caused by strong lightning is injected. Subsequently, a legality check is performed on the initial population, and cases such as landslides blocking the road surface and overly extreme wind speed settings that violate physical rationality are excluded. Finally, the valid scenario output is retained as the starting point for subsequent testing and evolution.

[0229] S4. Run the initial extreme driving case in the simulation environment: In S4, taking one of the scenarios with extreme weather and multiple interferences as an example, when the test vehicle enters the third lane at 88 km / h at 18:42, the simulation system synchronously triggers a three-level composite crisis event chain, including a sudden geological disaster 200 meters ahead that causes gravel to slide (the maximum rock diameter exceeds 40 cm and impacts the second lane at a speed of 12 m / s), a lateral displacement of 1.2 meters of the adjacent track freight train due to rail slippage that invades the roadway, and lightning interference that causes the in-vehicle vision system to fail for 300 milliseconds and the signal-to-noise ratio of the millimeter-wave radar to drop suddenly to 4 dB. At this time, the perception module completes the fusion and recognition of lidar point cloud and infrared camera data within about 430 milliseconds, detects 17 moving obstacles and the heat source signal of a breakdown vehicle 5 meters away, and the inertial navigation system calculates the road adhesion coefficient to be about 0.23 based on the tire slip rate.

[0230] The decision-making layer immediately calculates the dynamic risk map and determines that the collision probability of the emergency braking scheme is as high as 92% when the available braking distance is only 58 meters, and a lateral acceleration of up to 5.3 m / s 2 is required for a left lane change, while breaking through to the right road shoulder contains water eddies but the path curvature is relatively controllable. After comprehensive evaluation, the right obstacle avoidance strategy is selected.

[0231] The execution module implements four-wheel independent braking (the braking torque of the left front wheel can reach 1200 N·m) within 1.2 seconds and locks the rear axle differential. The vehicle cuts into the emergency lane with a yaw acceleration of 2.1 m / s 2 while maintaining a longitudinal acceleration of -4.5 m / s 2 to decelerate, and finally accurately avoids the obstacle at a distance of 1.8 meters from the obstacle, and the tire slip rate is always controlled near the 12% safety threshold.

[0232] S5. Evaluate the fitness of the extreme driving case based on the test result data: In this scenario, the environmental perception delay from rock fall detection to coordinate calibration is about 0.27 seconds. The collision risk assessment shows that if emergency braking is selected, the vehicle cannot stop within a distance of 58 meters. In terms of path planning error, the steering angle tracking deviation is 0.08 degrees and the wheel speed control accuracy is better than 0.3 km / h. The minimum dynamic safety distance of 1.5 meters (lower than the 2.0-meter collision alarm threshold of the system) triggers a secondary alarm. After comprehensively considering the perception accuracy, vehicle dynamic response robustness, and execution accuracy, calculate the comprehensive fitness value of this scenario and sort it with other scenarios. The overall test here shows that the perception module still maintains 89% environmental situation awareness integrity under strong lightning interference, but there is an 8% deviation in the estimation of the tire force saturation characteristic, indicating that there is still room for improvement on low-adhesion roads. It can be seen that this case has high-risk test value under multiple interferences.

[0233] S6. Further, based on the fitness evaluation results, perform selection, crossover, and mutation operations through a genetic algorithm: In the selection stage, screen out a number of cases with prominent extreme features or high collision risks as the candidate set according to the fitness threshold, and then perform pairwise crossover on key parameters such as road slope, rock mass sliding timing, and lightning intensity in the candidate set to generate a new batch of offspring scenarios. Randomly adjust the typhoon wind speed peak or rainfall extreme value, and change the occurrence time and duration of equipment failure through mutation operations to improve scenario diversity. Finally, merge the offspring cases after crossover and mutation with some parent cases to form the next-generation population of extreme driving cases.

[0234] The test log further shows that the timestamp synchronization accuracy of multi-sensor data decreases by 0.8 milliseconds under strong electromagnetic pulses, the estimation error of the planning module for tire force saturation reaches 8%, and the temperature of the hydraulic system rises to 92°C due to insufficient heat dissipation efficiency of the actuator during continuous emergency braking. Subsequently, a phase change material heat dissipation module will be installed to improve the safety margin under long-term composite working conditions.

[0235] S7. Termination of genetic algorithm iteration and post-analysis: Further, when the scenario fitness tends to converge or reaches the pre-set maximum number of rounds and scenario diversity threshold termination conditions after multiple rounds of iteration, the algorithm ends and outputs the final set of extreme driving cases. When summarizing and deeply analyzing all convergent scenarios, it is found that multiple lightning strikes interfere with the perception sensor, resulting in an approximately 15% extension of the fusion cycle. The vehicle dynamic model underestimates the adhesion characteristics of wet roads and needs to be further improved. Subsequently, a phase change material heat dissipation module needs to be considered to address the problem of excessive hydraulic temperature caused by continuous heavy braking. By comparing key indicators such as obstacle avoidance success rate, collision risk, and execution accuracy, accurately locate the shortcomings of the unmanned driving system in complex interference scenarios, provide a clear direction for continuously improving the perception fusion algorithm, vehicle dynamics control strategy, and hardware redundancy design, and also lay a relatively solid data-driven foundation for the safety guarantee in the subsequent real-road test session.

[0236] Through this embodiment, it can be seen that the method of the present invention has significantly improved in terms of the efficiency of generating extreme driving scenarios, the coverage rate of complex scenarios, and the accuracy and efficiency of unmanned driving system verification, providing reliable technical support for the practical application of unmanned driving technology.

[0237] The present invention realizes the dynamic optimization of extreme driving cases through a genetic algorithm, generates diverse extreme driving scenarios by combining fitness evaluation and evolutionary paths, can dynamically generate complex scenario parameter combinations through intelligent crossover and mutation operations, and at the same time achieves a more efficient coverage rate for low-probability and high-risk scenarios in the "long tail" problem. By introducing an adaptive perturbation mechanism, the simulation environment can generate complex conditions closer to actual driving situations, enhancing the flexibility and richness of scenario generation, and effectively making up for the deficiencies of the prior art in scenario coverage.

[0238] The present invention designs a multi-dimensional fitness evaluation index system, including the successful risk avoidance rate, path planning error, and vehicle dynamic response robustness, comprehensively analyzes the full-link performance of the unmanned driving system. The index system can simultaneously evaluate the collaborative effects of the perception, decision-making, and execution modules, provide more comprehensive performance evaluation data, adjust the influence of each evaluation index through a dynamic weight mechanism, optimize the rationality of fitness calculation, enable the rapid positioning and improvement of the performance bottleneck of the unmanned driving system, and effectively shorten the test and verification cycle.

[0239] The present invention supports modular design. By combining a parametric model and a simulation engine, it can achieve the controllable generation and accurate reproduction of extreme scenarios. Through parametric description and dynamic regulation, it can reproduce complex driving scenarios in the simulation environment at low cost and high efficiency, providing accurate data support for the algorithm debugging and performance optimization of the unmanned driving system. In addition, the introduction of an automated test framework greatly reduces manual intervention, ensuring the repeatability of test results and the objectivity of verification results.

[0240] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered by the protection scope of the present invention.

Claims

1. A method for generating and verifying extreme cases of driverless based on a simulation engine, characterized in that, It includes the following steps: S1. Construct a high-fidelity virtual simulation environment; S2. Initialize the parameterized model for generating extreme driving cases. The parameterized model is based on preset scenario rules, and the scenario rules are designed according to traffic regulations, actual driving behaviors, and known dangerous patterns; S3. Generate an initial population of extreme driving cases based on the genetic algorithm. The initial population contains multiple extreme driving cases. The parameters of the extreme driving cases are generated by random initialization or rule setting, and each extreme driving case is described by a set of scenario parameters; S4. Run the initial population of extreme driving cases in the simulation environment, conduct unmanned driving system tests on each extreme driving case, and collect the test result data of the unmanned driving system; S5. Conduct fitness evaluation on the population of extreme driving cases according to the test result data. The fitness evaluation is based on a multi-index system to quantify the test effect of each extreme driving case; S6. Based on the fitness evaluation results, perform selection, crossover, and mutation operations through the genetic algorithm. Select extreme driving cases with fitness higher than the threshold to enter the next generation. The crossover operation is used to generate new combinations of scenario parameters, and the mutation operation is used to introduce random perturbations to increase scenario diversity; S7. Conduct unmanned driving system tests on the evolved population of extreme driving cases in the simulation environment again, and repeat steps S4 to S6 until the preset termination conditions are reached. The termination conditions include that the scenario diversity reaches the target value or the fitness converges.

2. The method for generating and verifying an extreme case of driverless based on a simulation engine according to claim 1, wherein The above S1 includes the following specific steps: S11. Construct a road structure model, which includes straight roads, curved roads, intersections, and slope sections; S12. Construct a dynamic traffic scenario model, which is based on the kinematic characteristics and interaction rules of virtual vehicles. The interaction rules include following behavior, lane-changing behavior, and obstacle avoidance behavior; S13. Construct a physical interaction model, which is defined by simulating the dynamic interaction relationship between vehicles and the environment; S14. Set adjustable environmental parameters, including traffic flow density and the position and type of obstacles: The environmental parameters include weather conditions, defined as light intensity, rainfall, snowfall, and wind speed; Traffic flow density, defined as the flow density of vehicles per unit time; Obstacle position and type. The obstacle position is represented by coordinates (x o , y o ), and the obstacle types include static obstacles and dynamic obstacles; S15. Combine the road structure model, dynamic traffic scenario model, physical interaction model, and adjustable environmental parameters to generate a high-fidelity virtual simulation environment.

3. A method for generating and verifying unmanned driving extreme cases based on a simulation engine according to claim 1, characterized in that, The above S2 includes the following specific steps: S21. Initialize the initial state of the scenario, which is based on parameterized descriptions; Road geometric characteristic parameters (w d , r c , θ s , l d ), where w d represents the road width, r c represents the radius of curvature, θ s represents the slope angle, l d represents the road length; The position and status of traffic participants, defined as the initial position coordinates of a virtual vehicle Speed Acceleration and direction angle Environmental parameters, including weather conditions Among them represents the initial light intensity, represents the initial rainfall, represents the initial snowfall, represents the initial wind speed; S22. Define dynamic change rules, including vehicle kinematic dynamics: Among them, and represent the position of the vehicle at time t + 1, and Δt represents the time step; The environment changes dynamically, including the light intensity I w , the rainfall P r , the snowfall P s and the wind speed v w Rules of change over time: Where α and β represent the rates of dynamic change; I w (t) = I base * a weather Among them, I base is a local reference light intensity, and a weather is the weather attenuation coefficient; v w = v0 + Δv Where v0 is the basic wind speed and Δv is the wind speed amplitude. Obstacle behavior rules, defined as static obstacles keeping their positions unchanged and dynamic obstacles moving along specific paths; Among them, and represent the position of the dynamic obstacle at time t + 1, v o represents the obstacle speed, φ o represents the obstacle direction angle; S23. Set termination conditions: Scenario completion conditions, all traffic participants reach the target position or the simulation running time reaches the preset upper limit; System state conditions, the unmanned driving system has a collision, path planning failure, or timeout response; Under environmental stability conditions, the dynamic parameters reach a stable state or exceed the set threshold range. S24. Generate a parametric model based on the initial state of the scenario, the dynamic change rules, and the termination conditions.

4. A method for generating and verifying driverless extreme cases based on a simulation engine according to claim 1, characterized in that, The said S3 includes the following specific steps: S31. Define the initial population size N and population structure. The initial population consists of N extreme driving cases, and the parameters of each extreme driving case include road geometric characteristic parameters (w d , r c , θ s , l d ), traffic participant status parameters environmental parameters and obstacle parameters S32. Generate an initial population through random initialization. The random initialization is based on the set parameter range, and assigns random parameter values to each extreme driving case: Road width w d ∈ [3, 10] m, radius of curvature r c ∈ [10, 200] m, slope angle θ s ∈ [-15°, 15°], road length l d ∈ [50, 500] m; Virtual vehicle initial position coordinates m, initial speed m / s, initial acceleration m / s², initial direction angle radian; Illumination intensity Lumen, rainfall Millimeter / hour, snowfall Millimeter / hour, wind speed Meter / second; Initial position coordinates of the obstacle meters, with a speed of v o ∈ [0, 10] m / s, and a direction angle φ o ∈ [0, 2π] radians; S33. Generate specific extreme driving cases based on rule settings; S34. Conduct a legality check on the initial population, eliminate the extreme driving cases that do not meet the constraint conditions, and output the initial population of extreme driving cases that pass the legality check.

5. A method for generating and verifying driverless extreme cases based on a simulation engine according to claim 4, characterized in that, The generation of specific extreme driving cases based on rule settings includes the following categories: High-risk road condition rules: Curved road scenario: Set the radius of curvature r c ∈ [10, 50] meters to simulate a sharp turn scenario; Sloped road section scenario: Set the slope angle θ s ∈[-15°, -10°] ∪ [10°, 15°], to simulate the dynamic impact of uphill and downhill driving on driverless vehicles; Rainfall scenario: Set the rainfall amount mm / h², simulating sudden heavy rain; Wind speed scenario: Set the wind speed m / s² to simulate the impact of crosswind or tailwind on vehicle stability; Complex traffic flow scenario rules Multi-vehicle rear-end collision scenario: Set the initial position spacing d of virtual vehicles gap ∈ [0, 5] meters, and the speed difference v diff ∈ [5, 15] m / s to simulate a chain rear-end collision accident on a highway; Complex intersection interaction scenario: Set the initial coordinates of multiple vehicles converging at an intersection Speed m / s, direction angle Simulate the interaction dynamics in the case of signal light failure; Extreme weather condition rules: Snowstorm scenario: Set the snowfall amount mm / h 2, simulating the condition of reduced road surface skid resistance; Low visibility scenario: Set the light intensity in lumens to simulate the sensor's perception ability in haze weather; Sudden behavior rules of obstacles: Dynamic obstacle: Set the initial position of the obstacle Velocity v o ∈ [2, 8] m / s, direction angle φ o ∈ [0, π], simulating the scenario of pedestrians crossing the road Sudden obstacles: Dynamically generate obstacles, and the position change rule is: Among them, the speed v o ∈[5, 10] m / s, and the direction angle changes with time represents the perturbation angle; Vehicle dynamics failure scenario rules: Simulate the scenario of weakened tire grip: Set the friction coefficient μ ∈ [0.2, 0.4] to simulate the vehicle skidding caused by a wet and slippery road surface; Simulate emergency braking of the vehicle: Set the vehicle acceleration a v ∈[-10, -5] m / s² to evaluate the emergency response ability of the driverless system; Special scenario combination rules: Combine the dynamic traffic flow and extreme weather scenarios, and simulate the multi-vehicle rear-end collision in a heavy rain scenario; Combine the slope section and the dynamic obstacle scenario, and the impact of the sudden movement of the obstacle to the middle of the slope on the vehicle's decision-making path.

6. The method for generating and verifying driverless extreme cases based on a simulation engine according to claim 1, characterized in that The said S4 includes the following specific steps: S41. Load the initial population of extreme driving cases in a high-fidelity simulation environment. The parameters of each driving case include road geometric characteristic parameters, traffic participant state parameters, environmental parameters, and obstacle parameters; S42. Connect the unmanned driving system to the simulation environment and run each extreme driving case; S43. Collect the recognition accuracy η of the perception module during the operation of the driverless system p : The decision rationality of the path planning module is measured by the path error ∈ p : Response time τ of the execution module e , indicating the time delay from the input of the perception module to the vehicle's execution of an action; The vehicle state changes, including position, speed, acceleration, and steering angle.

7. A method for generating and verifying driverless extreme cases based on a simulation engine according to claim 1, characterized in that The said S5 includes the following specific steps: S51. Establish a multi-index system and calculate the following key evaluation indicators for each extreme driving case: Calculate the successful risk avoidance rate η by comprehensively considering the collision situation and environmental dynamic factors s : Among them, 1(d t ≤0) indicates whether a collision event occurs at time t, and d t is the minimum safety distance between the driverless vehicle and an obstacle or another vehicle, respectively represent the light intensity, rainfall, snowfall, and wind speed at time t, α1 is the environmental risk coefficient, and γ1 is the environmental disturbance weight; Collision count C sum , which is used to measure the cumulative situation of collision events in extreme driving cases: Path planning error ∈ C , which represents the average deviation between the actual driving trajectory of the driverless vehicle and the planned path. Referring to the form of path error and introducing the adaptive weight γ2 to consider the road geometric characteristic parameters (w d , r c , θ s , l d ): Among them, γ2 represents the modulation coefficient for the slope angle θ s and the radius of curvature r c ; Vehicle dynamic response robustness η r , which represents the ability of the vehicle to maintain stable driving under adverse environmental parameters, is defined by comprehensively considering sensor fluctuations and actuator module delays as: Among them, represents the response delay of the execution module at time t, represents the acceleration mutation amplitude of the vehicle at time t, and α2 and γ3 are the time-delay weight and acceleration perturbation weight of the dynamic response respectively; S52. Calculate the comprehensive fitness value F of each extreme driving case based on the multi - index system i : F i = w s ·η s - w c ·C sum - w p ·∈ C + w r ·η r ; Among them, w s , w c , w p , w r are the weighting coefficients of each index, and w s + w c + w p + w r = 1. The weighting coefficients can be configured according to the actual test requirements in different extreme scenarios S53. According to the comprehensive fitness value F i Sort the population of extreme driving cases, and use the sorting function Ψ(·) to assign indexes to all extreme driving cases: Among them, Sort represents sorting the fitness values of all extreme driving cases from high to low, and Ψ(F i ) is the ranking corresponding to the fitness value, and N c represents the total number of extreme driving cases participating in the evaluation.

8. A method for generating and verifying driverless extreme cases based on a simulation engine according to claim 1, characterized in that, The said S6 includes the following specific steps: S61. Determine the fitness threshold F th , and perform a selection operation based on the comprehensive fitness value F i to select all extreme driving cases that satisfy F i ≥F th to form the candidate set Ω1; S62. Randomly pair the extreme driving cases in the candidate set Ω1 pairwise and perform a crossover operation to generate new scenario parameter combinations; S63. Perform a mutation operation on the offspring scenario parameters generated after crossover, and introduce random perturbations to increase scenario diversity; S64. Combine the offspring extreme driving cases generated after the crossover and mutation operations with the parent extreme driving cases to form a new population, which is used as the input of the next generation of extreme driving cases.

Images