UC Irvine

Autonomous vehicles (AVs) are rapidly becoming a part of our everyday lives, and the number of AVs running on the road is growing faster. With the increased presence of AVs, there has been a rise in accidents, which underscores the critical need for enhanced safety and reliability in Autonomous Driving Software (ADS). Recent studies have proposed scenario-generation approaches to facilitate testing autonomous vehicles in broader scenarios. However, many of these approaches create scenarios that only run within a single ADS-Simulator environment, affecting the generalizability of the resulting approaches. One primary reason scenario generation approaches are limited to specific ADS-simulator environments is that each system uses distinct scenario and map formats. This format incompatibility limits the utility of scenarios that expose critical bugs, which is essential for improving ADS safety. To that end, we propose an ADS Scenario Transformer that enables transfer scenarios running on one ADS-Simulator to others. The transformation process targets two critical elements of a driving scenario: (1) the movement patterns of obstacles within the simulation, and (2) status changes of traffic signals. Our study investigates whether scenarios are transferred between different ADS-simulator environments while preserving original scenario information and explores differences in ADS behavior between original and transformed scenarios.