Automatic Driving Scene Sampling Method Based On Key Parameter Identification

Testing the safety of the autonomous driving system in dangerous scenarios can effectively judge its stability and adaptability and improve testing efficiency. Scene parameter sampling is one of the methods to generate dangerous scenes. However, because dangerous scenes are usually complex scenes, and complex scenes have the problem of too many parameters, the direct combination of parameters will produce a large number of safety scenes, resulting in the reduction of test efficiency. We propose a method to identify the critical scene parameters based on the above problems. The process can extract the essential parameters of complex scenes for scene sampling according to the different functions to be tested of the primary vehicle. We test complex overtaking-based scenarios against four sampling algorithms. Experiments show that identifying critical parameters in complex scenes can effectively improve the sampling efficiency of dangerous scenes.