Whitebox Testing, Debugging, and Repairing for Multi-module Autonomous Vehicles in Near-Collision Traffic Scenarios

Principal Investigator: Tianyi Zhang

This project seeks to develop principled algorithms and techniques for systematically testing, debugging, and repairing multi-module ADS to improve their safety and reliability. The core of
our research is (1) a new semantic-level Domain-Specific Language (DSL) to capture the richness of real-world traffic scenes and express rare, unexpected scenarios that may lead to collision (i.e., near-collision scenarios) and (2) a layered system abstraction that accounts for the interaction between different ADS modules and between DL models and logic-based code within a module. Building upon the DSL and layered abstraction, we will develop a new whitebox fuzz testing technique guided by a new hierarchical coverage metric, as well as new debugging and repair techniques that locate the unsafe ADS component and automatically fix it via data or code synthesis。

Personnel

Other PIs: Xiangyu Zhang

Students: Tu Zhi

Representative Publications

• Deng, Yao, Xi Zheng, Mengshi Zhang, Guannan Lou, and Tianyi Zhang. "Scenario-based test reduction and prioritization for multi-module autonomous driving systems." In Proceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, pp. 82-93. 2022.

• Lou, Guannan, Yao Deng, Xi Zheng, Mengshi Zhang, and Tianyi Zhang. "Testing of autonomous driving systems: where are we and where should we go?." In Proceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, pp. 31-43. 2022.