Online Pre-Training for Offline-to-Online Reinforcement Learning

TL;DR

This paper introduces OPT, a novel online pre-training method that enhances offline-to-online reinforcement learning by improving value estimation, leading to significant performance gains across various challenging environments.

Contribution

The paper proposes a new online pre-training approach that explicitly addresses value estimation issues in offline-to-online RL, improving fine-tuning effectiveness.

Findings

30% average performance improvement on D4RL benchmarks

Effective online pre-training enhances value estimation

Demonstrated on TD3 and SPOT algorithms

Abstract

Offline-to-online reinforcement learning (RL) aims to integrate the complementary strengths of offline and online RL by pre-training an agent offline and subsequently fine-tuning it through online interactions. However, recent studies reveal that offline pre-trained agents often underperform during online fine-tuning due to inaccurate value estimation caused by distribution shift, with random initialization proving more effective in certain cases. In this work, we propose a novel method, Online Pre-Training for Offline-to-Online RL (OPT), explicitly designed to address the issue of inaccurate value estimation in offline pre-trained agents. OPT introduces a new learning phase, Online Pre-Training, which allows the training of a new value function tailored specifically for effective online fine-tuning. Implementation of OPT on TD3 and SPOT demonstrates an average 30% improvement in performance across a wide range of D4RL environments, including MuJoCo, Antmaze, and Adroit.