Interactive Parallel Exploration for Reinforcement Learning in Continuous Action Spaces

In this paper, a new interactive parallel learning scheme is proposed to enhance the performance of off-policy continuous-action reinforcement learning. In the proposed interactive parallel learning scheme, multiple identical learners with their own value-functions and policies share a common experience replay buffer, and search a good policy in collaboration with the guidance of the best policy information. The information of the best policy is fused in a soft manner by constructing an augmented loss function for policy update to enlarge the overall search space by the multiple learners. The guidance by the previous best policy and the enlarged search space by the proposed interactive parallel learning scheme enable faster and better policy search in the policy parameter space. Working algorithms are constructed by applying the proposed interactive parallel learning scheme to several off-policy reinforcement learning algorithms such as the twin delayed deep deterministic (TD3) policy gradient algorithm and the soft actor-critic (SAC) algorithm, and numerical results show that the constructed IPE-enhanced algorithms outperform most of the current state-of-the-art reinforcement learning algorithms for continuous action control.