Physics-Guided Active Learning of Environmental Flow Fields

Reza Khodayi-mehr, Pingcheng Jian, Michael M. Zavlanos
Proceedings of The 5th Annual Learning for Dynamics and Control Conference, PMLR 211:928-940, 2023.

Abstract

We propose a physics-based method to learn environmental fields (EFs) using a mobile robot. Common data-driven methods require prohibitively many measurements to accurately learn such complex EFs. On the other hand, while physics-based models provide global knowledge of EFs, they require experimental validation, depend on uncertain parameters, and are intractable to solve onboard mobile robots. To address these challenges, we propose a Bayesian framework to select and improve upon the most likely physics-based models of EFs in real-time, from a pool of numerical solutions generated offline as a function of the uncertain parameters. Specifically, we use Gaussian Processes (GPs) to construct statistical models of EFs, and rely on the pool of numerical solutions to inform their prior mean. To incorporate flow measurements into these GPs, we control a custom-built mobile robot through a sequence of waypoints that maximize the information content of the measurements. We experimentally demonstrate that our proposed framework constructs a posterior distribution of the flow field that better approximates the real flow compared to the prior numerical solutions and purely data-driven methods.

Cite this Paper

BibTeX
@InProceedings{pmlr-v211-khodayi-mehr23a, title = {Physics-Guided Active Learning of Environmental Flow Fields}, author = {Khodayi-mehr, Reza and Jian, Pingcheng and Zavlanos, Michael M.}, booktitle = {Proceedings of The 5th Annual Learning for Dynamics and Control Conference}, pages = {928--940}, year = {2023}, editor = {Matni, Nikolai and Morari, Manfred and Pappas, George J.}, volume = {211}, series = {Proceedings of Machine Learning Research}, month = {15--16 Jun}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v211/khodayi-mehr23a/khodayi-mehr23a.pdf}, url = {https://proceedings.mlr.press/v211/khodayi-mehr23a.html}, abstract = {We propose a physics-based method to learn environmental fields (EFs) using a mobile robot. Common data-driven methods require prohibitively many measurements to accurately learn such complex EFs. On the other hand, while physics-based models provide global knowledge of EFs, they require experimental validation, depend on uncertain parameters, and are intractable to solve onboard mobile robots. To address these challenges, we propose a Bayesian framework to select and improve upon the most likely physics-based models of EFs in real-time, from a pool of numerical solutions generated offline as a function of the uncertain parameters. Specifically, we use Gaussian Processes (GPs) to construct statistical models of EFs, and rely on the pool of numerical solutions to inform their prior mean. To incorporate flow measurements into these GPs, we control a custom-built mobile robot through a sequence of waypoints that maximize the information content of the measurements. We experimentally demonstrate that our proposed framework constructs a posterior distribution of the flow field that better approximates the real flow compared to the prior numerical solutions and purely data-driven methods. } }
Endnote
%0 Conference Paper %T Physics-Guided Active Learning of Environmental Flow Fields %A Reza Khodayi-mehr %A Pingcheng Jian %A Michael M. Zavlanos %B Proceedings of The 5th Annual Learning for Dynamics and Control Conference %C Proceedings of Machine Learning Research %D 2023 %E Nikolai Matni %E Manfred Morari %E George J. Pappas %F pmlr-v211-khodayi-mehr23a %I PMLR %P 928--940 %U https://proceedings.mlr.press/v211/khodayi-mehr23a.html %V 211 %X We propose a physics-based method to learn environmental fields (EFs) using a mobile robot. Common data-driven methods require prohibitively many measurements to accurately learn such complex EFs. On the other hand, while physics-based models provide global knowledge of EFs, they require experimental validation, depend on uncertain parameters, and are intractable to solve onboard mobile robots. To address these challenges, we propose a Bayesian framework to select and improve upon the most likely physics-based models of EFs in real-time, from a pool of numerical solutions generated offline as a function of the uncertain parameters. Specifically, we use Gaussian Processes (GPs) to construct statistical models of EFs, and rely on the pool of numerical solutions to inform their prior mean. To incorporate flow measurements into these GPs, we control a custom-built mobile robot through a sequence of waypoints that maximize the information content of the measurements. We experimentally demonstrate that our proposed framework constructs a posterior distribution of the flow field that better approximates the real flow compared to the prior numerical solutions and purely data-driven methods.
APA
Khodayi-mehr, R., Jian, P. & Zavlanos, M.M.. (2023). Physics-Guided Active Learning of Environmental Flow Fields. Proceedings of The 5th Annual Learning for Dynamics and Control Conference, in Proceedings of Machine Learning Research 211:928-940 Available from https://proceedings.mlr.press/v211/khodayi-mehr23a.html.

Related Material