Sailing Towards Zero-Shot State Estimation using Foundation Models Combined with a UKF
paper-conference
Combining a foundation model pretrained on synthetic dynamical systems with the Unscented Kalman Filter for zero-shot state estimation.
Resources
DOI | arXiv | Preprint PDF | Code and data
Publication history
Preprint: 4 September 2025. Conference publication: IEEE CDC 2025 (9–12 December 2025).
Summary
This work combines a transformer foundation model pretrained on synthetic dynamical systems with an Unscented Kalman Filter. The foundation model supplies predictions of unfamiliar dynamics, while the filter incorporates measurements through a known sensor model. Evaluation on a containership benchmark tests zero-shot state estimation under varying dynamics. The paper also releases code and benchmark data for reproducing the experiments.
Citation
@inproceedings{holtmann2025sailing,
author = {Holtmann, Tobin and Stenger, David and Posada-Moreno, Andres and Solowjow, Friedrich and Trimpe, Sebastian},
booktitle = {2025 IEEE 64th Conference on Decision and Control (CDC)},
date = {2025},
doi = {10.1109/CDC57313.2025.11313025},
pages = {96--102},
title = {Sailing Towards Zero-Shot State Estimation using Foundation Models Combined with a UKF}
}