Attraction-Repulsion Actor-Critic for Continuous Control Reinforcement Learning
This addresses the challenge of learning in high-dimensional state spaces with deceptive rewards for reinforcement learning practitioners, though it appears incremental as it builds on population-based methods.
The paper tackles the problem of avoiding local optima in continuous control reinforcement learning by introducing a population-based approach that uses normalizing flows for attractive and repulsive operations between policies, achieving high performance gains on the MuJoCo suite compared to prior methods like Soft-Actor Critic.
Continuous control tasks in reinforcement learning are important because they provide an important framework for learning in high-dimensional state spaces with deceptive rewards, where the agent can easily become trapped into suboptimal solutions. One way to avoid local optima is to use a population of agents to ensure coverage of the policy space, yet learning a population with the "best" coverage is still an open problem. In this work, we present a novel approach to population-based RL in continuous control that leverages properties of normalizing flows to perform attractive and repulsive operations between current members of the population and previously observed policies. Empirical results on the MuJoCo suite demonstrate a high performance gain for our algorithm compared to prior work, including Soft-Actor Critic (SAC).