Virtual Test Platform Utilizing a Multi-agent Traffic Simulation to Develop a Sophisticated Automated Driving System

Abstract

This paper presents an effective testing methodology aimed at enhancing the development of a more advanced Automated Driving System (ADS). This is achieved through the generation of both intricate and challenging scenarios using a virtual test platform. Given the challenges associated with identifying and accumulating these scenarios via field operational tests, the a virtual test platform is expected to be a valuable approach for enhancing the performance of ADS. However, thea virtual test platform must provide appropriate scenarios in accordance with the current performance capabilities of the respective ADS. To address this, the current research has realized a virtual test platform by integrating a multi-agent traffic simulation with ADS. This integration incorporates functionalities that enable ADS to effectively navigate complex and perplexing scenario. The results of this study reveal that these functionalities not only facilitate the extraction of rare scenarios but also to reinforce significant diversification required to develop sophisticated ADS.