Policy Transfer via Latent Graph Planning
Keywords: Reinforcement Learning, Transfer Learning
Abstract: We introduce a transfer learning framework for deep reinforcement learning that integrates graph-based planning with self-supervised representation learning to efficiently transfer knowledge across tasks. While standard reinforcement learning aims to learn policies capable of solving long-horizon tasks, the resulting policies often fail to generalize to novel tasks and environments. Our approach addresses this limitation by decomposing long-horizon tasks into sequences of transferable short-horizon tasks modeled by goal-conditioned policies. We utilize a planning graph to generate fine-grained sub-goals that guide these short-horizon policies to solve novel long-horizon tasks. Experimental results show that our method improves sample efficiency and demonstrates an improved ability to solve sparse-reward and long-horizon tasks compared to baseline methods in challenging single-agent and multi-agent scenarios. In particular, compared to the state-of-the-art, our method achieves the same or better expected policy reward while requiring fewer training samples when learning novel tasks.
Primary Area: transfer learning, meta learning, and lifelong learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4203
Loading