A win for EpiSci and national defense: Prashant Ganesh, Principal Research Engineer at EpiSci, co-authored research introducing a novel approach that combines hybrid control theory with reinforcement learning to enhance the safety and reliability of RL policies in complex environments. When asked about the project, he explained, "MultiHyRL is a cutting-edge algorithm that strengthens existing reinforcement learning models, making them more reliable even in challenging, unpredictable environments." #MultiHyRL #DefenseTech #AIEngineering
Professor and Chair of Electrical and Computer Engineering, Director of the Cyber-Physical Systems Research Center, Director of the CITRIS Aviation Initiative, Consultant
Today, we presented our paper titled "MultiHyRL: Robust Hybrid RL for Obstacle Avoidance against Adversarial Attacks on the Observation Space," co-authored with Zachary Bell from AFRL, Prashant Ganesh from EpiSci, and my Ph.D. student Jan de Priester at the 2024 Reinforcement Learning Conference (RLC24). The paper proposes a new hybrid RL algorithm featuring hysteresis-based switching to guarantee robustness against adversarial attacks on the observation space for vehicles operating in environments with multiple obstacles. We demonstrate the robustness of our approach in various challenging obstacle avoidance settings where Proximal Policy Optimization (PPO), a traditional RL method, fails. #RLC2024 #HybridSystems Baskin Engineering at UCSC