Locally Weighted Regression Pseudo-Rehearsal for Adaptive Model Predictive Control

Grady R. Williams, Brian Goldfain, Keuntaek Lee, Jason Gibson, James M. Rehg, Evangelos A. Theodorou
Proceedings of the Conference on Robot Learning, PMLR 100:969-978, 2020.

Abstract

We consider the problem of online adaptation of a neural network designed to represent system dynamics. The neural network model is intended to be used by an MPC control law for autonomous control. This problem is challenging because both input and target distributions are non-stationary, and naive approaches to online adaptation result in catastrophic forgetting. We present a novel online learning method, which combines the pseudo-rehearsal method with locally weighted projection regression. We demonstrate the effectiveness of the resulting Locally Weighted Projection Regression Pseudo-Rehearsal (LW-PR2) method on an autonomous vehicle in simulation and real world data collected with a 1/5 scale autonomous vehicle.

Cite this Paper

BibTeX
@InProceedings{pmlr-v100-williams20a, title = {Locally Weighted Regression Pseudo-Rehearsal for Adaptive Model Predictive Control}, author = {Williams, Grady R. and Goldfain, Brian and Lee, Keuntaek and Gibson, Jason and Rehg, James M. and Theodorou, Evangelos A.}, booktitle = {Proceedings of the Conference on Robot Learning}, pages = {969--978}, year = {2020}, editor = {Kaelbling, Leslie Pack and Kragic, Danica and Sugiura, Komei}, volume = {100}, series = {Proceedings of Machine Learning Research}, month = {30 Oct--01 Nov}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v100/williams20a/williams20a.pdf}, url = {https://proceedings.mlr.press/v100/williams20a.html}, abstract = {We consider the problem of online adaptation of a neural network designed to represent system dynamics. The neural network model is intended to be used by an MPC control law for autonomous control. This problem is challenging because both input and target distributions are non-stationary, and naive approaches to online adaptation result in catastrophic forgetting. We present a novel online learning method, which combines the pseudo-rehearsal method with locally weighted projection regression. We demonstrate the effectiveness of the resulting Locally Weighted Projection Regression Pseudo-Rehearsal (LW-PR2) method on an autonomous vehicle in simulation and real world data collected with a 1/5 scale autonomous vehicle.} }
Endnote
%0 Conference Paper %T Locally Weighted Regression Pseudo-Rehearsal for Adaptive Model Predictive Control %A Grady R. Williams %A Brian Goldfain %A Keuntaek Lee %A Jason Gibson %A James M. Rehg %A Evangelos A. Theodorou %B Proceedings of the Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2020 %E Leslie Pack Kaelbling %E Danica Kragic %E Komei Sugiura %F pmlr-v100-williams20a %I PMLR %P 969--978 %U https://proceedings.mlr.press/v100/williams20a.html %V 100 %X We consider the problem of online adaptation of a neural network designed to represent system dynamics. The neural network model is intended to be used by an MPC control law for autonomous control. This problem is challenging because both input and target distributions are non-stationary, and naive approaches to online adaptation result in catastrophic forgetting. We present a novel online learning method, which combines the pseudo-rehearsal method with locally weighted projection regression. We demonstrate the effectiveness of the resulting Locally Weighted Projection Regression Pseudo-Rehearsal (LW-PR2) method on an autonomous vehicle in simulation and real world data collected with a 1/5 scale autonomous vehicle.
APA
Williams, G.R., Goldfain, B., Lee, K., Gibson, J., Rehg, J.M. & Theodorou, E.A.. (2020). Locally Weighted Regression Pseudo-Rehearsal for Adaptive Model Predictive Control. Proceedings of the Conference on Robot Learning, in Proceedings of Machine Learning Research 100:969-978 Available from https://proceedings.mlr.press/v100/williams20a.html.

Related Material