Abstract: It’s a critical step to generate test cases covering a wide range of high-risk driving scenarios in Autonomous Vehicle (AV) simulation tests. This study aims to propose a method to generate multi-dimensional cut-in high-risk scenarios. A total of 405 high-risk cut-in events were extracted from the Shanghai Naturalistic Driving Study (NDS). Four parameters at the initial moment were used to characterize the cut-in scenarios, which can better reproduce the entire cut-in process. The Gibbs sampling method and the Latin hypercube sampling (LHS) method were used to generate multi-dimensional cut-in high-risk scenarios. Root Mean Square Percentage Error (RMSPE) is used to measure the distribution difference between the NDS data and the sampled scenario set. The results showed that the Gibbs sampling method could better reproduce the original NDS data distribution by yielding a lower RMSPE than that of the LHS method. At the same time, the performance of the Gibbs sampling method to generate multi-dimensional cut-in high-risk scenarios is better than LHS, which can better support autonomous driving testing.
Shuang Liu, Xuesong Wang*, Xiaoyan Xu, Kun Xie. Multi-dimensional Cut-in High-risk Test Scenarios Generation Method for Autonomous Vehicles. Transportation Research Board 101st Annual Meeting, Washington D.C., USA, 2022. 1.9-13.