Method for extracting hazardous scenario
By determining functional scenarios and simulation parameters in the autonomous driving system, filtering out parameter groups that meet preset conditions for simulation testing, the problem of low risk scenario capture rate in the existing technology is solved, and more efficient verification of autonomous driving algorithms is achieved.
Patent Information
- Application Number
- PCT/CN2024/094960
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-29
- Filing Date
- 2024-05-23
- Publication Date
- 2024-12-05
AI Technical Summary
It is difficult for existing autonomous driving systems to fully cover various dangerous scenarios in road testing, resulting in waste of test resources and safety risks, and it is difficult for existing methods to effectively capture extreme dangerous scenarios.
By selecting functional scenarios, we determine the simulation parameter types and value ranges, establish constraints, generate alternative parameter groups, and conduct tests in the simulation environment to filter out the risk evaluation value that meets the preset conditions, thereby extracting the hazardous scenarios.
It improves the probability and efficiency of capturing dangerous scenarios, reduces the consumption of test resources, and provides a more economical and safe verification method.
Smart Images

Figure CN2024094960_05122024_PF_FP_ABST
Abstract
Description
A method for extracting dangerous scenes Technical Field
[0001] The present invention relates to the technical field of unmanned driving test technology, and in particular to a method for extracting dangerous scenes. Background Art
[0002] With the rapid development of autonomous driving technology, the demand for testing autonomous driving systems is growing. The RAND Corporation indicates that an autonomous driving system requires at least 11 billion miles of testing before it can be ready for mass production. However, it's difficult to fully enumerate all dangerous scenarios, and the number of dangerous and extreme scenarios doesn't increase with increasing test mileage. Therefore, in-vehicle testing is prone to inadequate scenario coverage, and these omissions pose safety risks to autonomous vehicles.
[0003] In the era of software-defined cars, a mature autonomous driving system is portable and updatable. An intelligent driving system encompasses perception, decision-making, and execution. The execution component requires connectivity with the vehicle's various actuation systems, including the by-wire chassis, steering, and braking systems. Many automotive electronics suppliers have already achieved significant success in this area. The 11 billion-mile road test program emphasizes the fact that the primary challenges in achieving autonomous driving lie in perception and decision-making. These two components are crucial to autonomous driving systems. These components can be integrated or implemented independently. Consequently, numerous simulation testing platforms have emerged. Developers input perception and decision-making algorithms into the platform, which then integrates them into a virtual vehicle. The virtual vehicle is tested in a variety of complex simulation scenarios, providing feedback on the algorithm's performance and functionality. This feedback allows for identification of deficiencies and targeted algorithm improvements. These tests require the creation of as many dangerous and extreme scenarios as possible, encompassing real-world driving scenarios, to screen the autonomous driving algorithms.
[0004] Dangerous scenarios (pre-collision scenarios) are the processes before a vehicle collides with a dynamic or static element. The classification of collisions can be used to better understand the scenarios and the causes of their occurrence. The current classification standards for dangerous scenarios mainly include the 37 dangerous operating conditions of the National Highway Traffic Safety Administration (NHTSA) and the accident types of the US Highway Safety Administration shown in the SHRP2 (STRATEGIC HIGHWAY RESEARCH PROGRAM) report. This type of scenario is a challenging test scenario without a collision. Testing dangerous scenarios can significantly shorten the verification time of autonomous driving algorithm testing and is an economically feasible and safe verification method.
[0005] Currently, some autonomous driving manufacturers primarily conduct road testing. However, relying solely on road testing incurs significant human, material, and time costs, and the limited hazardous conditions it can cover make it difficult to capture the majority of dangerous scenarios. Furthermore, existing scenarios are mostly simple and repetitive, with a large number of natural scenarios being non-critical, and the incidence of extremely dangerous scenarios being very low. Using test scenario parameter combinations can identify dangerous scenarios with low occurrence rates, but managing the sheer volume of test cases resulting from these parameter combinations and evaluating the safety metrics for these dangerous scenarios remain challenges.
[0006] Summary of the Invention
[0007] The purpose of the present invention is to provide a method for extracting dangerous scenes in order to overcome the defects of the above-mentioned prior art, so as to solve or partially solve the problems that the probability of capturing dangerous scenes by the existing method is low and the extraction process consumes a lot of resources.
[0008] The purpose of the present invention can be achieved by the following technical solutions:
[0009] The present invention provides a method for extracting dangerous scenes, comprising the following steps:
[0010] Select a functional scenario for testing, and determine the types of simulation parameters required to construct the functional scenario and the value range of each simulation parameter;
[0011] Establish constraints on simulation parameters based on the characteristics of the selected functional scenario;
[0012] Based on the value range and constraint conditions of the simulation parameters, discretization is performed at preset parameter intervals to obtain multiple candidate parameter groups;
[0013] A test environment is constructed based on the candidate parameter group, a hazard evaluation value is obtained through simulation testing, and the candidate parameter group whose hazard evaluation value meets preset conditions is saved to realize the extraction of dangerous scenarios.
[0014] As a preferred technical solution, the process of obtaining multiple candidate parameter groups includes the following steps:
[0015] Based on the simulation parameters and the parameter intervals, parameter samples are obtained through exhaustive enumeration, and parameter samples that meet the constraint conditions are retained as the candidate parameter group.
[0016] As a preferred technical solution, the risk assessment value includes the time required for the test vehicle to collide with the target and / or the maximum deceleration.
[0017] As a preferred technical solution, the preset conditions are specifically: the time required for the test vehicle to collide with the target is between 0.5s and 1.5s and / or the maximum deceleration is less than 3m·s-2 .
[0018] As a preferred technical solution, the types of simulation parameters include one or more of the speed of the main vehicle, the speed of the other vehicle, the lane change time, the width of the road, the longitudinal distance between the main vehicle and the other vehicle, the overlap rate, and the offset rate.
[0019] As a preferred technical solution, the constraints include that the main vehicle's speed is greater than or equal to the speed of the other vehicle in front, the overlap rate is within 50% to 100%, and the other vehicle's offset rate is within 0% to 50%.
[0020] As a preferred technical solution, the parameter interval is specifically:
[0021] If the parameter type is the speed interval of the main vehicle, the parameter interval is 5-20 km·h -1 If the parameter type is acceleration / deceleration, the parameter interval is 0.5-2m·s -2 , if the parameter type is simulation initial position spacing, the parameter interval is 2-10m.
[0022] As a preferred technical solution, the functional scenarios include forward collision warning and lane change warning.
[0023] As a preferred technical solution, the process of building a test environment based on the candidate parameter group includes the following steps:
[0024] The alternative parameter groups are imported into the simulation software in a parallel manner to realize the construction of the test environment.
[0025] As a preferred technical solution, the simulation software is VTD, PreScan or CarMaker.
[0026] Compared with the prior art, the present invention has the following advantages:
[0027] (1) Improving the probability of capturing dangerous scenarios: This method first selects the functional scenario to be tested, determines the type and value range of the required simulation parameters according to the characteristics of the functional scenario, and establishes constraint relationships. A large number of alternative parameter groups are generated at preset intervals and screened using constraint relationships. The alternative parameter groups are used to perform simulation tests in a simulation environment to obtain hazard evaluation values. The alternative parameter groups that meet the preset rules are retained, thereby realizing the extraction of dangerous scenarios. Compared with the exhaustive method, this method limits the value range according to the characteristics of the functional scenario before generating the alternative parameter groups, and then uses constraint relationships to screen the alternative parameter groups. This effectively reduces the test volume before actual testing, thereby improving the probability of capturing dangerous scenarios.
[0028] (2) The preset conditions and parameter interval value ranges are provided, which further improves the probability of obtaining dangerous scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] FIG1 is a flow chart of a method for extracting dangerous scenes in Example 1. DETAILED DESCRIPTION
[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.
[0031] Example 1
[0032] As shown in Figure 1, this embodiment provides a method for extracting dangerous scenarios. This method reduces the number of samples required to find dangerous scenario test cases by constraining relevant parameters. The method then clusters the discretized scenario parameters, combines the input parameter values of interest to the test vehicle, and generates a large number of test samples through an exhaustive method. Finally, the combined parameters are imported into the autonomous driving simulation software to quickly and efficiently extract dangerous scenarios. This method specifically includes the following steps:
[0033] Step S1: Design a functional scenario based on the requirements of the test algorithm, such as forward collision warning, lane change warning, etc., list the parameters required for the scenario, and the process of establishing the scenario itself is a process of adjusting parameters, such as the speed of the host vehicle, the speed of the other vehicle, the lane change time, the width of the road, the acceleration / deceleration of the host vehicle, the acceleration / deceleration of the other vehicle, the type of the host vehicle, the type of the other vehicle, the longitudinal distance between the host vehicle and the other vehicle, the overlap rate, the offset rate, etc.
[0034] Step S1, refine the required scenario and set the range of required parameter changes, such as determining the speed range and acceleration range of the host vehicle and other vehicles. At the same time, parameters such as weather and lighting can also be included. Parameters outside this range are invalid parameters and cannot achieve the purpose of verifying the autonomous driving algorithm.
[0035] Step S2 constrains the scenario parameters themselves through specific constraints, such as speed and acceleration. For example, in a collision scenario, if the main vehicle's speed is slower than that of the other vehicle ahead, a collision will not occur, further reducing the number of samples. Furthermore, collision scenario constraints also include an overlap ratio of 50% to 100% and an offset ratio of the other vehicle within 0% to 50%.
[0036] Step S3: discretize the reduced parameters, for example, discretize the speed of the main vehicle by 10 km / h. -1 Discretize, the other car has acceleration, constant speed, and deceleration driving states, and the acceleration and deceleration are divided into 1m·s -2Discretize, discretize the initial positions of the two at intervals of 5 m, etc. After clustering, further analyze and group the parameter combinations that may produce dangerous scenarios together.
[0037] Step S4, the combined parameters design the required test scenarios through an exhaustive method, perform permutations and combinations within their respective different intervals, generate a large number of sample test scenarios, and import them into simulation software through parallel computing, such as VTD, PreScan, CarMaker, etc.
[0038] Step S5, TTC (Time-to-collision or time-to-contact) is the time required for the host vehicle to collide with the target. Based on TTC and the maximum deceleration, perform scenario screening. When TTC < 0.5 s, the collision probability is 100%, and when TTC ≥ 2.5 s, the collision probability is 0. Here, select the scenarios where 0.5 s < TTC ≤ 1.5 s as dangerous scenarios. If the deceleration of the host vehicle is too large, it will greatly affect the safety of the vehicle and the comfort of the passengers. Here, set the maximum value of the deceleration to 3 m·s -2 ². And compare the magnitude relationship between the preset value and the actual value of the dangerous scenario. Define TTC and the maximum deceleration as the methods for evaluating test scenarios, and select the scenarios that satisfy 0.5 s < TTC ≤ 1.5 s and the maximum deceleration is less than 3 m·s⁻² as valuable dangerous scenarios, and verify the scenarios through the software-in-the-loop method.
[0039] Step S6, if TTC and the maximum deceleration are within the preset range, then this scenario is a dangerous scenario that can be used for verifying the autonomous driving algorithm, and the autonomous driving algorithm can be further improved based on this; if not within the preset range, this scenario is not an optimal test scenario.
[0040] Step S7, determine the relevant parameters that pass the evaluation and establish valuable dangerous scenarios.
[0041] This method reduces the sample size by constraining relevant parameters, reduces the sample size of finding test cases for dangerous scenarios, and increases the probability of capturing dangerous scenarios.
[0042] Embodiment 2
[0043] This embodiment provides an electronic device, including: one or more processors and a memory. The memory stores one or more programs, and the one or more programs include instructions for executing the method for extracting dangerous scenarios as described in Embodiment 1.
[0044] Embodiment 3
[0045] This embodiment provides a computer-readable storage medium, including one or more programs for execution by one or more processors of an electronic device, wherein the one or more programs include instructions for executing the dangerous scenario extraction method as described in Example 1.
[0046] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.
Claims
1. A method for extracting dangerous scenes, characterized in that: The steps include: Select a functional scenario for testing, and determine the type of simulation parameters required to construct the functional scenario and the value range of each simulation parameter; Establish constraints on simulation parameters based on the characteristics of the selected functional scenario; Based on the value range and constraint conditions of the simulation parameters, discretization is performed with preset parameter intervals to obtain multiple candidate parameter groups; A test environment is constructed based on the candidate parameter group, a hazard evaluation value is obtained through simulation testing, and the candidate parameter group whose hazard evaluation value meets preset conditions is saved to realize the extraction of dangerous scenes.
2. The method for extracting dangerous scenes according to claim 1, characterized in that: The process of obtaining multiple candidate parameter groups includes the following steps: Based on the simulation parameters and the parameter intervals, parameter samples are obtained by exhaustive enumeration, and parameter samples that meet the constraint conditions are retained as the candidate parameter group.
3. The method for extracting dangerous scenes according to claim 1, characterized in that: The risk evaluation value includes the time required for the test vehicle to collide with the target and / or the maximum deceleration.
4. The method for extracting dangerous scenes according to claim 1, characterized in that: The preset conditions are: the time required for the test vehicle to collide with the target is between 0.5s and 1.5s and / or the maximum deceleration is less than 3m·s -2 .
5. The method for extracting dangerous scenes according to claim 1, characterized in that: The types of simulation parameters include one or more of the speed of the main vehicle, the speed of the other vehicle, the lane change time, the width of the road, the acceleration / deceleration of the main vehicle, the acceleration / deceleration of the other vehicle, the type of the main vehicle, the type of the other vehicle, the longitudinal distance between the main vehicle and the other vehicle, the overlap rate, and the offset rate.
6. The method for extracting dangerous scenes according to claim 1, characterized in that: The constraint conditions include that the speed of the main vehicle is greater than or equal to the speed of the other vehicle in front, the overlap rate is within 50% to 100%, and the offset rate of the other vehicle is within 0% to 50%.
7. The method for extracting dangerous scenes according to claim 1, characterized in that: The parameter interval is specifically: If the parameter type is the speed interval of the main vehicle, the parameter interval is 5-20km·h -1 , if the parameter type is For acceleration and deceleration, the parameter interval is 0.5-2m·s -2 , if the parameter type is simulation initial position spacing, the parameter interval is 2-10m.
8. The method for extracting dangerous scenes according to claim 1, characterized in that: The functional scenarios described include forward collision warning and lane change warning.
9. The method for extracting dangerous scenes according to claim 1, characterized in that: The process of building a test environment based on the candidate parameter group includes the following steps: The alternative parameter groups are imported into the simulation software in a parallel manner to realize the construction of the test environment.
10. The method for extracting dangerous scenes according to claim 9, characterized in that: The simulation software is VTD, PreScan or CarMaker.
Citation Information
Patent Citations
Virtual test scene construction method and device
CN114139329A
Method for automatically generating specific scene based on constraint model
CN114492008A
Automatic driving decision-making dangerous scene generation method, system, equipment and medium
CN115795808A
Dangerous scene extraction method
CN116663282A
Generating Accurate and Diverse Simulations for Evaluation of Autonomous-Driving Systems
US20220198096A1
