UC Irvine

Autonomous vehicles (AVs) are increasingly prevalent in our daily routines, with examples such as robotaxis and robot deliveries becoming more commonplace. This trend emphasizes the importance of thorough testing procedures to guarantee their safety and dependability. Autonomous driving software (ADS) enables AVs to navigate using complex algorithms and machine learning techniques. While field operational tests are regularly employed to assess ADS functionality, they are hindered by cost and geographic limitations. Virtual testing emerges as a safer and more controlled alternative. Past research has primarily focused on validating the approach on Baidu Apollo, with limited exploration of another open-source ADS, Autoware. This raises questions about the generalizability of existing methodologies across different ADS systems. To address this gap, this study aims to transfer scenoRITA to generate safety-critical and motion sickness-inducing test scenarios for Autoware. Our empirical results reveal that scenoRITA identifies 63 unique safety and comfort violations in Autoware.