Development of Simulation-Based Testing Scenario Generator for Robustness Verification of Autonomous Vehicles

Abstract

This paper explores how to effectively increase the testing scenarios for robustness verification of autonomous vehicles by adjusting various traffic scenarios and different severity of quantifiable weather and interference parameters. The automated test and verification tools developed are based on the VTD vehicle simulation platform, which can quickly generate various weather conditions. In addition, considering the camera disturbed by the raindrops on windshield, we develop a faster and more realistic raindrop interference generation methodology, which can quantify the generation of different raindrop size, density, and flow interference, to enhance the realism and diversity of the testing scenarios of autonomous perception system. We demonstrate the usage of the proposed tools to generate the different testing scenarios under various weather conditions and interference. Then, the AI object detection models were tested under the generated testing scenarios to investigate the effect of raindrops on windshield on the robustness of the object detection model. The contribution of this work is to propose an effective simulation-based testing scenario generator to increase the test coverage and shorten the verification time of the robustness test of perception systems.