Swapped goal-conditioned offline reinforcement learning

TL;DR

This paper introduces a goal-swapping technique and a new offline RL method, DQAPG, to improve generalization and performance in goal-conditioned tasks, especially in complex manipulation scenarios.

Contribution

The paper proposes a novel goal-swapping data augmentation method and the DQAPG algorithm, enhancing offline GCRL performance and robustness against noise and extrapolation errors.

Findings

DQAPG outperforms state-of-the-art methods on benchmark tasks.

Goal-swapping improves test results in goal-conditioned offline RL.

The method achieves success on complex in-hand manipulation tasks.

Abstract

Offline goal-conditioned reinforcement learning (GCRL) can be challenging due to overfitting to the given dataset. To generalize agents' skills outside the given dataset, we propose a goal-swapping procedure that generates additional trajectories. To alleviate the problem of noise and extrapolation errors, we present a general offline reinforcement learning method called deterministic Q-advantage policy gradient (DQAPG). In the experiments, DQAPG outperforms state-of-the-art goal-conditioned offline RL methods in a wide range of benchmark tasks, and goal-swapping further improves the test results. It is noteworthy, that the proposed method obtains good performance on the challenging dexterous in-hand manipulation tasks for which the prior methods failed.