Research And Application Of Test Scenario Generation Method For Autonomous Driving Vehicles

With the development of autonomous driving,how to verify the safety of autonomous vehicles has become more and more important.Scenario-based testing is one of the important methods of autonomous driving testing.In order to quickly obtain smallprobability test cases,this paper applies the combinatorial testing method and Bayesian network to the field of test scenario generation,and proposes a fast method of generating test cases with specified probability.The generation of test scenarios consists of two parts,scenario analysis and test case generation.Among them,scenario analysis refers to determining the elements of the scenario and constructing the parameter space and constraint set.In this article,the scenario is simplified to consist of environmental elements,road elements,and object elements.On this basis,the parameter space is determined by combining the driving area and driving tasks.The constraints between the scenario parameters are transformed into forbidden tuples,and the constraint simplification and implicit constraint search are performed on them,and finally a complete set of constraints is obtained.The structure of the Bayesian network is established by referring to the relative relationship between the objects in the scenario,and the conditional probability value is obtained by mapping the relative parameters of the objects.In the test case generation process,the probability of the test case and the coverage of the uncovered combination set are weighted,and the selection parameters with different values are calculated.The expected parameters and adjustment parameters in the calculation can be modified to achieve generation test cases with different probabilities.Based on the existing test case generation algorithms AETG and CTBC,the parameter value indicators are replaced with selection parameters,so that the probability of the generated test case set is controllable.Finally,the method is applied to the automated test of the automatic parking.The generated test cases are automatically converted into scenarios in Pre Scan,and the simulation results are scored according to the current automatic parking test standards.Comparing the scores of test case sets with different probabilities,it is found that the algorithm under test has the lowest score and the most failures in the small probability test case set,which verifies the effectiveness of the method proposed in this paper.