POMDIFFUSER: LONG-MEMORY MEETS LONG- PLANNING FOR POMDPS
Keywords: Reinforcement learning, Partial observability, Long memory, Planning
Abstract: Effective long-term planning in complex environments benefits from not only leveraging immediate information but also utilizing past experiences. Drawing inspiration from how humans use long-term memory in decision-making, we propose the POMDiffuser framework, an approach to planning in partially observable environments. While conventional Diffuser models often memorize specific environments, POMDiffuser explores the potential of learning to plan from memory, with the aim of generalizing to new scenarios. By incorporating a memory mechanism in POMDP scenarios, our model extends diffusion-based planning models into the realm of meta-learning with carefully designed tasks that require the diffusion planner to demonstrate both long-term planning and memory utilization. We investigated existing diffusion-based models, focusing on their applicability, computational efficiency, and performance trade-offs.
Primary Area: reinforcement 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: 10128
Loading