Pretraining Representations for Data-Efficient Reinforcement Learning

TL;DR

This paper introduces a pretraining method for reinforcement learning that uses unlabeled data to improve data efficiency, achieving near-human performance on Atari with limited interaction data.

Contribution

The authors propose a novel pretraining approach combining latent dynamics modeling and goal-conditioned RL, significantly enhancing data efficiency in deep reinforcement learning.

Findings

Outperforms prior offline pretraining methods on Atari with 100k steps

Approaches human-level performance with larger models and diverse data

Requires substantially less data than previous methods

Abstract

Data efficiency is a key challenge for deep reinforcement learning. We address this problem by using unlabeled data to pretrain an encoder which is then finetuned on a small amount of task-specific data. To encourage learning representations which capture diverse aspects of the underlying MDP, we employ a combination of latent dynamics modelling and unsupervised goal-conditioned RL. When limited to 100k steps of interaction on Atari games (equivalent to two hours of human experience), our approach significantly surpasses prior work combining offline representation pretraining with task-specific finetuning, and compares favourably with other pretraining methods that require orders of magnitude more data. Our approach shows particular promise when combined with larger models as well as more diverse, task-aligned observational data -- approaching human-level performance and data-efficiency on Atari in our best setting. We provide code associated with this work at https://github.com/mila-iqia/SGI.