Marginalized State Distribution Entropy Regularization in Policy Optimization
This work addresses the challenge of state space exploration for reinforcement learning agents, particularly in sparse reward and continuous control domains, representing an incremental improvement over existing entropy regularization methods.
The authors tackled the problem of insufficient state space exploration in reinforcement learning by proposing entropy regularization based on the marginal state distribution, which achieved superior state space coverage on gridworld domains and empirical gains in sparse reward 3D maze navigation and continuous control tasks.
Entropy regularization is used to get improved optimization performance in reinforcement learning tasks. A common form of regularization is to maximize policy entropy to avoid premature convergence and lead to more stochastic policies for exploration through action space. However, this does not ensure exploration in the state space. In this work, we instead consider the distribution of discounted weighting of states, and propose to maximize the entropy of a lower bound approximation to the weighting of a state, based on latent space state representation. We propose entropy regularization based on the marginal state distribution, to encourage the policy to have a more uniform distribution over the state space for exploration. Our approach based on marginal state distribution achieves superior state space coverage on complex gridworld domains, that translate into empirical gains in sparse reward 3D maze navigation and continuous control domains compared to entropy regularization with stochastic policies.