Closed-Loop Simulation Test Method for Autonomous Vehicles Based on Adversarial Scenarios

Abstract

Using simulation technology to verify the safety of autonomous vehicles (AVs) is a critical aspect of the development of autonomous driving technology. The primary objective is to reduce the time and cost of AV testing while enhancing testing efficiency. In this study, we propose a closed-loop simulation test method for AVs based on an adversarial scenario generation process. A comprehensive closed-loop simulation system was established, incorporating Simulink, Prescan, Carsim, and hardware-in-the-loop systems.To swiftly identify challenging test scenarios, we formulated an optimization model for safety evaluation, which is based on an improved driving safety field model. We employed the genetic algorithm to effectively solve this optimization problem, allowing for the rapid generation of difficult scenarios.Experimental results demonstrate that our proposed method can effectively generate challenging scenarios, enabling the swift confirmation of the safety performance boundaries of the system under test. This approach significantly improves the overall testing efficiency and provides valuable insights for the advancement of autonomous driving technologies.