A joint imitation-reinforcement learning framework for reduced baseline regret

摘要: In various control task domains, existing controllers provide a baseline level of performance that—though possibly suboptimal—should be maintained. Reinforcement learning (RL) …