FALCON: Fourier Adaptive Learning and Control for Disturbance Rejection Under Extreme Turbulence
-
1.
California Institute of Technology
-
2.
Nvidia (United States)
Abstract
Controlling aerodynamic forces in turbulent conditions is crucial for UAV operation. Traditional reactive methods often struggle due to unpredictable flow and sensor noise. We present FALCON (Fourier Adaptive Learning and Control), a model-based reinforcement learning framework for effective modeling and control of aerodynamic forces under turbulent flows. FALCON leverages two key insights: turbulent dynamics are well-modeled in the frequency domain, and most turbulent energy is concentrated in low-frequencies. FALCON learns a concise Fourier basis to model system dynamics from 35 s of flow data. To address sensor limitations, FALCON models dynamics using a short history of actions and measurements. With this approach, FALCON applies model predictive control for safe and efficient control. Tested in the Caltech wind tunnel under highly turbulent conditions, FALCON learns to control the underlying nonlinear dynamics with less than 9 min of data, consistently outperforming state-of-the-art methods. We provide guarantees for FALCON, ensuring stability and robustness.
Copyright and License
This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Acknowledgement
The authors would like to thank Professor John Dabiri (Caltech) for the insightful discussion regarding this work. Anima Anandkumar is Bren chair and Schmidt sciences AI 2050 senior fellow and this work was supported by the Center for Autonomous Systems and Technologies at Caltech, the National Science Foundation Graduate Research Fellowship under Grant No. DGE-1745301, the US Office of Naval Research (MURI grant N00014-18-12624).
Contributions
All the authors conceptualized and planned the testing methodology together. S.L. and P.R. were responsible for the algorithm design, prototype design, manufacturing, data acquisition, programming, visualizations, theoretical derivations, and writing the original paper drafts. K.A., B.H., M.G., and A.A. supervised and provided advice throughout the entire project, and helped edit and review the paper.
Files
Additional details
- Caltech Center for Autonomous Systems and Technologies
- National Science Foundation
- Graduate Research Fellowship DGE-1745301
- Office of Naval Research
- Multidisciplinary University Research Initiative (MURI) N00014-18-12624
- Accepted
-
2024-08-28Accepted
- Available
-
2024-09-24Published online
- Publication Status
- Published