Project Grant 2240981

The Regents of the University of Michigan received a $480,000 project grant from the National Science Foundation under the Engineering federal grant program (CFDA 47.041) to develop foundational advances in robust reinforcement learning solutions and safe, constrained reinforcement learning methods with provable guarantees. The research will focus on advancing algorithmic solutions for reinforcement learning-based control in cyber-physical systems, using smart traffic signal control systems as the motivating application. Key activities include developing robust reinforcement learning incorporating model knowledge, off-policy methods to address distributional shifts, and efficient constrained reinforcement learning algorithms. The algorithms will be evaluated using traffic signal control simulations. Broader impacts include curriculum development, student recruitment, and research outreach. The three-year award period began June 1, 2023. No subaward details were provided.