EMI: Exploration with Mutual Information

TL;DR

EMI is an exploration method in reinforcement learning that uses embedding representations of states and actions to guide exploration through forward prediction, improving performance on sparse reward tasks.

Contribution

It introduces a novel embedding-based exploration approach that does not rely on generative decoding, enhancing exploration efficiency in sparse reward environments.

Findings

Competitive results on locomotion tasks

Effective on image-based Atari exploration tasks

Does not depend on generative decoding of full observations

Abstract

Reinforcement learning algorithms struggle when the reward signal is very sparse. In these cases, naive random exploration methods essentially rely on a random walk to stumble onto a rewarding state. Recent works utilize intrinsic motivation to guide the exploration via generative models, predictive forward models, or discriminative modeling of novelty. We propose EMI, which is an exploration method that constructs embedding representation of states and actions that does not rely on generative decoding of the full observation but extracts predictive signals that can be used to guide exploration based on forward prediction in the representation space. Our experiments show competitive results on challenging locomotion tasks with continuous control and on image-based exploration tasks with discrete actions on Atari. The source code is available at https://github.com/snu-mllab/EMI .