Excellence in Research - Research of Swarm Control by Synthesizing Reinforcement Learning with Temporal Logic in Swarm Robotics

Alabama A&M University, Normal AL

Investigators

Abstract

This project delves into the evolving realm of swarm robotics using integrated reinforcement learning and temporal logic. The focus is on the collaborative dynamics of multi-robot systems, specifically Unmanned Aerial Vehicles (UAVs), to navigate and operate in uncharted environments safely. The endeavor seeks to bridge the gap between high-level behavioral specifications and the robots' operational actions through an innovative integration of Reinforcement Learning (RL) and Linear Temporal Logic (LTL). This research stands at the forefront of addressing the pressing need for methodologies that can accurately specify, verify, and control the collective behaviors of robotic swarms in real-time. The broader implications of this project underscore its significance in promoting safer, more efficient UAV deployments in complex scenarios, highlighting the project's alignment with the National Science Foundation's mission to advance the frontiers of science and engineering. Additionally, the project's commitment to involving students from Historically Black Colleges and Universities (HBCUs) underscores its dedication to diversifying the STEM fields, fostering inclusivity and innovation in technology. The goal of the project is to build a framework that facilitates the design, development, and validation of a synthesized control policy allowing swarm UAVs to coordinate with each other to achieve a mission successfully in an unknown environment. By leveraging the synergies between RL and LTL, the project introduces a novel approach to optimizing reward functions and control policies, ensuring robust adaptability and efficient coordination among UAVs. The technical framework encompasses theoretical modeling, simulation-based testing, and real-world validation on a sophisticated swarm robotics platform. The project meticulously outlines a lifecycle that spans from the synthesis of RL and TL for dynamic behavior specification, through the iterative learning and refinement of swarm behavior, to the ultimate validation and analysis of performance in real-world settings. This comprehensive approach not only promises significant advancements in swarm robotics control but also sets a precedent for future research in multi-agent systems and dynamic operational environments. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.