BeTrusty: Real-World Autonomous Agents through Trustworthy Reinforcement Learning

BeTrusty aims to develop trustworthy autonomous agents capable of operating safely and effectively in real-world applications. The project addresses key limitations of reinforcement learning (RL) by developing multi-agent cooperation methods based on high-level actions (macro-actions) and formal verification techniques to provide safety guarantees during both training and deployment. The resulting methods will be integrated into a unified framework for trustworthy reinforcement learning and validated in two application domains: autonomous warehouse robotics and power grid operations.