PhD Dissertation Defense by CHENG Mingfei | Automated Testing and Enhancement for Autonomous Driving Systems

ABOUT THE TALK

Autonomous Driving Systems (ADSs) determine how autonomous vehicles understand and respond to traffic, making their trustworthiness increasingly important as deployment expands. This dissertation advances ADS assurance from automated testing to adaptive runtime enhancement through four connected studies. The first study addresses repetitive safety-critical failure discovery by characterizing temporal ego-vehicle behavior patterns and combining behavioral-diversity feedback with violation-oriented guidance. This approach generates critical scenarios associated with distinct driving behaviors. The second study evaluates path-planning robustness using non-invasive mutations that preserve the original optimal path. A consistency-based oracle identifies cases in which an ADS remains safe but selects a non-optimal path. The third study examines cooperation among ADS-controlled vehicles by formalizing deadlocks as persistent circular wait-for relations and combining spatio-temporal feedback with intersection-oriented mutation to generate deadlock scenarios. The final study explores adaptive runtime enhancement by mitigating potentially unsafe decisions while limiting unnecessary intervention. Together, these studies address failure diversity, decision robustness, multi-vehicle cooperation, and runtime risk reduction, providing practical methods and empirical insights for developing more trustworthy ADSs.

ABOUT THE SPEAKER

CHENG Mingfei is a PhD candidate in Computer Science at Singapore Management University, supervised by Assistant Professor XIE Xiaofei. He received his M.S. degree from Beijing University of Posts and Telecommunications (BUPT) and a joint B.S. degree from BUPT and Queen Mary University of London. He was also a visiting researcher at the University of Ottawa. His research focuses on improving the reliability and trustworthiness of AI-enabled systems through software engineering (SE) approaches, particularly testing, repair, and runtime monitoring. His primary research interests include autonomous driving systems and LLM agent systems. He has published more than 15 papers in leading SE conferences and journals, including ICSE, ISSTA, FSE, ASE, TSE, and TOSEM. He also contributes to the SE research community as a program committee member and reviewer for conferences and journals including ICSE, ASE, ISSTA, TSE, and TOSEM. He has received the SMU Presidential Doctoral Fellowship and the ICSE 2026 Shadow Program Committee Distinguished Reviewer Award.