Abstract

The performance of off-policy learning, including deep Q-learning and deep deterministic policy gradient (DDPG), critically depends on the choice of the exploration strategy. Existing exploration methods are mostly based on adding noises to the on-going actor policy and therefore only explore locally close to what the actor policy dictates. In this work, we develop a simple meta-policy gradient algorithm that allows us to adaptively learn the exploration policy in DDPG. Our algorithm allows us to train flexible exploration behaviors that are independent of the actor policy, yielding a more global exploration that significantly accelerates Q-learning. With an extensive study, we show that our method significantly improves the sample-efficiency of DDPG on a variety of reinforcement learning continuous control tasks.