Stein Variational Goal Generation for adaptive Exploration in Multi-Goal Reinforcement Learning

TL;DR

This paper introduces Stein Variational Goal Generation (SVGG), a method that adaptively generates goals of appropriate difficulty for multi-goal reinforcement learning, improving exploration and generalization in challenging environments.

Contribution

SVGG leverages Stein Variational Gradient Descent to sample goals of intermediate difficulty based on a learned model of the agent's capabilities, enhancing exploration in sparse reward settings.

Findings

SVGG outperforms state-of-the-art methods in success coverage on hard exploration tasks.

SVGG demonstrates robustness and recovery when environmental conditions change.

The approach effectively balances goal difficulty to facilitate learning.

Abstract

In multi-goal Reinforcement Learning, an agent can share experience between related training tasks, resulting in better generalization for new tasks at test time. However, when the goal space has discontinuities and the reward is sparse, a majority of goals are difficult to reach. In this context, a curriculum over goals helps agents learn by adapting training tasks to their current capabilities. In this work we propose Stein Variational Goal Generation (SVGG), which samples goals of intermediate difficulty for the agent, by leveraging a learned predictive model of its goal reaching capabilities. The distribution of goals is modeled with particles that are attracted in areas of appropriate difficulty using Stein Variational Gradient Descent. We show that SVGG outperforms state-of-the-art multi-goal Reinforcement Learning methods in terms of success coverage in hard exploration problems, and demonstrate that it is endowed with a useful recovery property when the environment changes.