A novel framework for promoting safe reinforcement learning by distilling offline expert demonstrations into lightweight, safe policy networks.
Jan 15, 2024
