Project Grant 2231257

This $413,694 project grant from the National Science Foundation's Computer and Information Science and Engineering program will support the development of formal methods for safe, efficient, and transferable autonomy in cyber-physical systems driven by deep reinforcement learning controllers. Over a three-year period from April 2023 through March 2026, researchers at Washington University in University City, Missouri will pursue three research thrusts: 1) accelerated and safe reinforcement learning for temporal logic control objectives; 2) safe transfer learning for temporal logic control objectives; and 3) compositional verification of temporal logic properties for cyber-physical systems with neural network controllers. The research leverages tools from formal methods, machine learning, and control theory with the goal of enabling safety-critical applications for cyber-physical systems in areas such as environmental monitoring, infrastructure inspection, autonomous driving, and healthcare. Validation and demonstration activities are planned using mobile aerial and ground robots performing autonomous surveillance, delivery, and mobile manipulation tasks.