What is Essential for Unseen Goal Generalization of Offline Goal-conditioned RL?
This addresses the challenge of training agents to achieve out-of-distribution goals from offline datasets, which is crucial for real-world applications where goals may not be seen during training, representing an incremental advance in offline GCRL.
The paper tackles the problem of unseen goal generalization in offline goal-conditioned reinforcement learning (GCRL), identifying key factors through theoretical and empirical analysis, and proposes a new method (GOAT) that outperforms state-of-the-art methods by a large margin on a benchmark with 9 IID and 17 OOD tasks.
Offline goal-conditioned RL (GCRL) offers a way to train general-purpose agents from fully offline datasets. In addition to being conservative within the dataset, the generalization ability to achieve unseen goals is another fundamental challenge for offline GCRL. However, to the best of our knowledge, this problem has not been well studied yet. In this paper, we study out-of-distribution (OOD) generalization of offline GCRL both theoretically and empirically to identify factors that are important. In a number of experiments, we observe that weighted imitation learning enjoys better generalization than pessimism-based offline RL method. Based on this insight, we derive a theory for OOD generalization, which characterizes several important design choices. We then propose a new offline GCRL method, Generalizable Offline goAl-condiTioned RL (GOAT), by combining the findings from our theoretical and empirical studies. On a new benchmark containing 9 independent identically distributed (IID) tasks and 17 OOD tasks, GOAT outperforms current state-of-the-art methods by a large margin.