Time-aware World Model: Adaptive Learning of Task Dynamics
Keywords: RL, Dynamics, World Model
TL;DR: Modeling Dynamic Systems for World Models
Abstract: In this work, we introduce Time-Aware World Model, a model-based approach designed to explicitly incorporate the temporal dynamics of environments. By conditioning on the time step size, $\Delta t$, and training over a diverse range of $\Delta t$ values - rather than relying on a fixed time step size - our model enables learning of both high- and low-frequency task dynamics in real-world control problems. Inspired by the information-theoretic principle that the optimal sampling rate varies depending on the underlying dynamics of different physical systems, our time-aware model enhances both performance and learning efficiency. Empirical evaluations demonstrate that our model consistently outperforms baseline approaches across different observation rates in various control tasks, using the same number of training samples and iterations. We will release our source code on GitHub once the final review decisions are made.
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: 13660
Loading