Stabilizing Off-Policy Q-Learning via Bootstrapping Error Reduction

TL;DR

This paper identifies bootstrapping error as a key instability in off-policy Q-learning and proposes BEAR, a new algorithm that constrains action selection to improve robustness when learning from fixed off-policy data.

Contribution

The paper provides a theoretical analysis of bootstrapping error and introduces BEAR, a novel method to mitigate this error by constraining action selection during off-policy learning.

Findings

BEAR learns robustly from various off-policy distributions.

Theoretical analysis links bootstrapping error to action selection outside data distribution.

Empirical results show improved stability and performance on continuous control tasks.

Abstract

Off-policy reinforcement learning aims to leverage experience collected from prior policies for sample-efficient learning. However, in practice, commonly used off-policy approximate dynamic programming methods based on Q-learning and actor-critic methods are highly sensitive to the data distribution, and can make only limited progress without collecting additional on-policy data. As a step towards more robust off-policy algorithms, we study the setting where the off-policy experience is fixed and there is no further interaction with the environment. We identify bootstrapping error as a key source of instability in current methods. Bootstrapping error is due to bootstrapping from actions that lie outside of the training data distribution, and it accumulates via the Bellman backup operator. We theoretically analyze bootstrapping error, and demonstrate how carefully constraining action selection in the backup can mitigate it. Based on our analysis, we propose a practical algorithm, bootstrapping error accumulation reduction (BEAR). We demonstrate that BEAR is able to learn robustly from different off-policy distributions, including random and suboptimal demonstrations, on a range of continuous control tasks.