Alessio Gambi //

Beyond Scenario-based Testing of Single Ego Vehicles
ABSTRACT-Current approaches for automatically testing Autonomous Driving Systems (ADS) in simulations generate diverse and challenging driving scenarios. However, they are fundamentally limited in several aspects, including challenging a single ego vehicle at the time using pre-programmed Non-playable Characters (NPC) and using existing or simple flat maps. Testing the ego-vehicle against only pre-programmed NPCs generates many irrelevant, i.e., non-bug-revealing test cases, whereas using only existing or simple flat maps neglects to test the AVs under diverse environmental conditions. In this talk, I presented recent results pushing state-of-the-art ADS Testing beyond single-ego vehicle testing, including a novel approach to testing ADS interactions (joint work with Paolo Arcaini, NII, Japan) and an approach leveraging large language models (LLM) to generate three-dimensional virtual roads from descriptions in natural language (joint work with IMC University of Applied Sciences Krems and BeamNG GmbH).
BIO-Dr. Alessio Gambi holds a permanent scientist position at the Austrian Institute of Technology, the largest RTO in Austria. Before joining AIT, he was a senior lecturer at IMC University of Applied Sciences Krems, a postdoc at Uni Passau, CISPA, and TUWien, and a PhD student at USI Lugano. His research lies at the intersection of Software Engineering, Software Testing, and Autonomous Systems. Currently, he is investigating approaches to improving the quality of autonomous driving systems. Besides this, he is interested in Computer Science Education and how to improve the teaching of Software Testing using Gamification and Large Language Models.
ℹ️ si.usi.ch/semi...