Generalization Generation of Hazardous Lane-changing Scenarios for Automated Vehicle Testing

To address the issue of hazardous lane-changing scenario construction in automated vehicle virtual testing, proposed a data-model-driven method for generally producing hazardous lane-changing scenarios. Based on emergency lane-changing data in NGSIM US101 Dataset, an emergency lane-changing trajectories producing method called batch normalization-attention mechanism-sequence generative adversarial nets (BN-AM-SeqGAN) with policy gradient is proposed based on sequence generative adversarial network. The safety distance based constraint model for two vehicle lane-changing statesis built, and the general approach of producing hazardous lane-changing test scenarios is designed. The library of hazardous lane-changing test scenarios is finally achieved. According to the experimental findings, the root mean square error of the lane-changing completion time distribution for produced 50 000 emergency trajectories is 0.63. Among the 50 000 generated hazardous lane-changing scenarios, the collision time between the tested automated vehicle and the lane-changing background vehicle is less than 1 s in 99.54% of the scenarios. Results show that the proposed method can effectively produce hazardous lane-changing scenarios for automated vehicle testing. © 2023 Science Press. All rights reserved.