Robust Regression for Safe Exploration in Control

Anqi Liu, Guanya Shi, Soon-Jo Chung, Anima Anandkumar, Yisong Yue
Proceedings of the 2nd Conference on Learning for Dynamics and Control, PMLR 120:608-619, 2020.

Abstract

We study the problem of safe learning and exploration in sequential control problems. The goal is to safely collect data samples from operating in an environment, in order to learn to achieve a challenging control goal (e.g., an agile maneuver close to a boundary). A central challenge in this setting is how to quantify uncertainty in order to choose provably-safe actions that allow us to collect informative data and reduce uncertainty, thereby achieving both improved controller safety and optimality. To address this challenge, we present a deep robust regression model that is trained to directly predict the uncertainty bounds for safe exploration. We derive generalization bounds for learning and connect them with safety and stability bounds in control. We demonstrate empirically that our robust regression approach can outperform the conventional Gaussian process (GP) based safe exploration in settings where it is difficult to specify a good GP prior.

Cite this Paper

BibTeX
@InProceedings{pmlr-v120-liu20a, title = {Robust Regression for Safe Exploration in Control}, author = {Liu, Anqi and Shi, Guanya and Chung, Soon-Jo and Anandkumar, Anima and Yue, Yisong}, booktitle = {Proceedings of the 2nd Conference on Learning for Dynamics and Control}, pages = {608--619}, year = {2020}, editor = {Bayen, Alexandre M. and Jadbabaie, Ali and Pappas, George and Parrilo, Pablo A. and Recht, Benjamin and Tomlin, Claire and Zeilinger, Melanie}, volume = {120}, series = {Proceedings of Machine Learning Research}, month = {10--11 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v120/liu20a/liu20a.pdf}, url = {https://proceedings.mlr.press/v120/liu20a.html}, abstract = {We study the problem of safe learning and exploration in sequential control problems. The goal is to safely collect data samples from operating in an environment, in order to learn to achieve a challenging control goal (e.g., an agile maneuver close to a boundary). A central challenge in this setting is how to quantify uncertainty in order to choose provably-safe actions that allow us to collect informative data and reduce uncertainty, thereby achieving both improved controller safety and optimality. To address this challenge, we present a deep robust regression model that is trained to directly predict the uncertainty bounds for safe exploration. We derive generalization bounds for learning and connect them with safety and stability bounds in control. We demonstrate empirically that our robust regression approach can outperform the conventional Gaussian process (GP) based safe exploration in settings where it is difficult to specify a good GP prior.} }
Endnote
%0 Conference Paper %T Robust Regression for Safe Exploration in Control %A Anqi Liu %A Guanya Shi %A Soon-Jo Chung %A Anima Anandkumar %A Yisong Yue %B Proceedings of the 2nd Conference on Learning for Dynamics and Control %C Proceedings of Machine Learning Research %D 2020 %E Alexandre M. Bayen %E Ali Jadbabaie %E George Pappas %E Pablo A. Parrilo %E Benjamin Recht %E Claire Tomlin %E Melanie Zeilinger %F pmlr-v120-liu20a %I PMLR %P 608--619 %U https://proceedings.mlr.press/v120/liu20a.html %V 120 %X We study the problem of safe learning and exploration in sequential control problems. The goal is to safely collect data samples from operating in an environment, in order to learn to achieve a challenging control goal (e.g., an agile maneuver close to a boundary). A central challenge in this setting is how to quantify uncertainty in order to choose provably-safe actions that allow us to collect informative data and reduce uncertainty, thereby achieving both improved controller safety and optimality. To address this challenge, we present a deep robust regression model that is trained to directly predict the uncertainty bounds for safe exploration. We derive generalization bounds for learning and connect them with safety and stability bounds in control. We demonstrate empirically that our robust regression approach can outperform the conventional Gaussian process (GP) based safe exploration in settings where it is difficult to specify a good GP prior.
APA
Liu, A., Shi, G., Chung, S., Anandkumar, A. & Yue, Y.. (2020). Robust Regression for Safe Exploration in Control. Proceedings of the 2nd Conference on Learning for Dynamics and Control, in Proceedings of Machine Learning Research 120:608-619 Available from https://proceedings.mlr.press/v120/liu20a.html.

Related Material