Learning-based Safe Control under Uncertainty

Safety is a primary requirement for many autonomous systems. For example, automated vehicles should be at least as safe as human-driven vehicles. Ensuring safety is challenging due to inevitable uncertainties and task complexity.

The scientific problem in this project is how to automatically synthesize safe controllers for uncertain systems under complex tasks. Solving this problem is of great importance for both academic and industrial areas.

The scientific novelty of this project lies in developing a new interdisciplinary approach based on the unique concept of temporal logic trees that I recently proposed, and in integrating learning and reachability analysis.

Based on the proposed approach, there will be three workpackages with different emphases and achievable tasks for a total grant period of three years. The preliminary results from my previous research provide solid foundation for this project.

During the grant period, I will be hosted at two complementary departments of top ranking in the fields and will be supervised by two highly renowned professors whose research interests are closely relevant to the project.

The theoretical contributions of the project will provide safe guarantees for autonomous systems under complex tasks while the new technological tools developed will contribute to accelerating the deployment of safety-critical systems in real world. All the results from this project will significantly advance the frontiers of research on safe control.