Bayesian Learning of Adaptive Koopman Operator with Application to Robust Motion Planning for Autonomous Trucks
Keywords: Koopman Theory, Motion Planning, Autonomous Systems
TL;DR: A Bayesian approach to learning Koopman operators is introduced to tackle uncertainty estimation and temporal shifts in dynamical systems, allowing for fast, accurate, and dynamically aware motion planning.
Abstract: Koopman theory has recently been shown to enable an efficient data-driven approach for modeling physical systems, offering a linear framework despite underlying nonlinear dynamics. It is, however, not clear how to account for uncertainty or temporal distributional shifts within this framework, both commonly encountered in real-world autonomous driving with changing weather conditions and time-varying vehicle dynamics. In this work, we introduce BLAK, Bayesian Learning of Adaptive Koopman operator to address these limitations. Specifically, we propose a Bayesian Koopman operator that incorporates uncertainty quantification, enabling more robust predictions. To tackle distributional shifts, we propose an online adaptation mechanism, ensuring the operator remains responsive to changes in system dynamics. Additionally, we apply the architecture to motion planning and show that it gives fast and precise predictions. By leveraging uncertainty awareness and real-time updates, our planner generates dynamically accurate trajectories and makes more informed decisions. We evaluate our method on real-world truck dynamics data under varying weather conditions—such as wet roads, snow, and ice—where uncertainty and dynamic shifts are prominent, as well as in other simulated environments. The results demonstrate our method’s ability to deliver accurate, uncertainty-aware open-loop predictions for dynamic systems.
Supplementary Material: zip
Primary Area: applications to robotics, autonomy, planning
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: 11900
Loading