Quality Assurance Automation in Autonomous Systems

Robots and autonomous systems are finding their way to interact with the public and failures in these systems could be extremely expensive, even deadly. However, low-cost software-based simulation could be a promising approach to systematically test robotics systems and prevent failures as early as possible. In our early work, we showed that the majority of bugs could actually be reproduced and discovered using low-fidelity simulation environment. We created a high-level framework for automated testing of popular ARDUPILOT systems. In this work, I propose novel approaches to automatically infer powerful representation of system models, and generate test suites with the purpose of enhancing automated fault localization performance and describing the root cause of failures. Finally, I propose to use those novel approaches to inform the construction of automated program repair techniques for autonomous systems.