Maximum Entropy Gain Exploration for Long Horizon Multi-goal Reinforcement Learning

TL;DR

This paper introduces a novel exploration strategy for long-horizon multi-goal reinforcement learning that maximizes the entropy of achieved goals, significantly improving sample efficiency in complex tasks.

Contribution

It proposes a new intrinsic goal-setting method based on entropy maximization of achieved goals, enhancing exploration in long-horizon multi-goal tasks.

Findings

Achieves an order of magnitude better sample efficiency than previous methods.

Effective in maze navigation and block stacking tasks.

Focuses exploration on the frontier of achievable goal sets.

Abstract

What goals should a multi-goal reinforcement learning agent pursue during training in long-horizon tasks? When the desired (test time) goal distribution is too distant to offer a useful learning signal, we argue that the agent should not pursue unobtainable goals. Instead, it should set its own intrinsic goals that maximize the entropy of the historical achieved goal distribution. We propose to optimize this objective by having the agent pursue past achieved goals in sparsely explored areas of the goal space, which focuses exploration on the frontier of the achievable goal set. We show that our strategy achieves an order of magnitude better sample efficiency than the prior state of the art on long-horizon multi-goal tasks including maze navigation and block stacking.