Efficient Deep Reinforcement Learning Through Policy Transfer

Bibtex

This paper proposes a novel Policy Transfer Framework (PTF) to accelerate RL by adaptively reusing knowledge from source policies of relevant tasks. Our framework learns when and which source policy is the best to reuse for the target policy and when to terminate it by modeling multi-policy transfer as the option learning problem. PTF can be easily combined with existing deep RL approaches. Experimental results show it significantly accelerates the learning process and surpasses state-of-the-art policy transfer methods in terms of learning efficiency and final performance in both discrete and continuous action spaces.