Jueming Hu "Reinforcement Learning for Planning and Optimization in Aviation Systems"

10:00 am
Friday, February 9, 2024
270 NCB Hall
Faculty Host: Dr. Rakesh Kapania

Abstract: Unmanned Aircraft Systems (UAS) are increasingly integrating into today’s airspace, providing societal benefits across civil and military applications. Despite the success of Artificial Intelligence, particularly Reinforcement Learning (RL), in domains like gameplay and robotics, challenges arise when applying RL to safety-critical systems, including UAS, due to data efficiency and constraint satisfaction issues. In this seminar, Dr. Hu will present weaving together theories from RL, formal methods, and optimization to address these challenges, thereby enhancing the safety, efficiency, and reliability of UAS. In the first part of the seminar, she will introduce the implementation of the RL algorithm for obstacle avoidance. The proposed model is able to provide accurate and robust guidance and resolve conflicts with a success rate of over 99%. Moving forward, the second part of the seminar introduces the incorporation of formal methods into RL, specifically reward machines, to enable agents to efficiently accomplish temporally extended tasks. In the third part, Dr. Hu explores the utilization of prior knowledge of optimization constraints to obtain optimal results that satisfy the constraints, with application to maintenance scheduling to avoid failures and achieve the optimal business reward.

Bio: Jueming Hu is currently a Postdoctoral Researcher in the School for Engineering of Matter, Transport, and Energy at Arizona State University, where she works on jointly learning high-level knowledge and optimal policies in Reinforcement Learning.  Jueming received her Ph.D. and master’s degrees from Arizona State University in 2023 and 2018, respectively. She obtained her bachelor’s degree from Southeast University in China in 2017. Her research interests include reinforcement learning, UAS traffic management, and optimization.