Agile Catching with Whole-Body MPC and Blackbox Policy Learning

Saminda Abeyruwan, Alex Bewley, Nicholas Matthew Boffi, Krzysztof Marcin Choromanski, David B D’Ambrosio, Deepali Jain, Pannag R Sanketi, Anish Shankar, Vikas Sindhwani, Sumeet Singh, Jean-Jacques Slotine, Stephen Tu
Proceedings of The 5th Annual Learning for Dynamics and Control Conference, PMLR 211:851-863, 2023.

Abstract

We address a benchmark task in agile robotics: catching objects thrown at high-speed. This is a challenging task that involves tracking, intercepting, and cradling a thrown object with access only to visual observations of the object and the proprioceptive state of the robot, all within a fraction of a second. We present the relative merits of two fundamentally different solution strategies: (i) Model Predictive Control using accelerated constrained trajectory optimization, and (ii) Reinforcement Learning using zeroth-order optimization. We provide insights into various performance tradeoffs including sample efficiency, sim-to-real transfer, robustness to distribution shifts, and whole-body multimodality via extensive on-hardware experiments. We conclude with proposals on fusing "classical" and "learning-based" techniques for agile robot control.

Cite this Paper

BibTeX
@InProceedings{pmlr-v211-abeyruwan23a, title = {Agile Catching with Whole-Body MPC and Blackbox Policy Learning}, author = {Abeyruwan, Saminda and Bewley, Alex and Boffi, Nicholas Matthew and Choromanski, Krzysztof Marcin and D'Ambrosio, David B and Jain, Deepali and Sanketi, Pannag R and Shankar, Anish and Sindhwani, Vikas and Singh, Sumeet and Slotine, Jean-Jacques and Tu, Stephen}, booktitle = {Proceedings of The 5th Annual Learning for Dynamics and Control Conference}, pages = {851--863}, 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/abeyruwan23a/abeyruwan23a.pdf}, url = {https://proceedings.mlr.press/v211/abeyruwan23a.html}, abstract = {We address a benchmark task in agile robotics: catching objects thrown at high-speed. This is a challenging task that involves tracking, intercepting, and cradling a thrown object with access only to visual observations of the object and the proprioceptive state of the robot, all within a fraction of a second. We present the relative merits of two fundamentally different solution strategies: (i) Model Predictive Control using accelerated constrained trajectory optimization, and (ii) Reinforcement Learning using zeroth-order optimization. We provide insights into various performance tradeoffs including sample efficiency, sim-to-real transfer, robustness to distribution shifts, and whole-body multimodality via extensive on-hardware experiments. We conclude with proposals on fusing "classical" and "learning-based" techniques for agile robot control. } }
Endnote
%0 Conference Paper %T Agile Catching with Whole-Body MPC and Blackbox Policy Learning %A Saminda Abeyruwan %A Alex Bewley %A Nicholas Matthew Boffi %A Krzysztof Marcin Choromanski %A David B D’Ambrosio %A Deepali Jain %A Pannag R Sanketi %A Anish Shankar %A Vikas Sindhwani %A Sumeet Singh %A Jean-Jacques Slotine %A Stephen Tu %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-abeyruwan23a %I PMLR %P 851--863 %U https://proceedings.mlr.press/v211/abeyruwan23a.html %V 211 %X We address a benchmark task in agile robotics: catching objects thrown at high-speed. This is a challenging task that involves tracking, intercepting, and cradling a thrown object with access only to visual observations of the object and the proprioceptive state of the robot, all within a fraction of a second. We present the relative merits of two fundamentally different solution strategies: (i) Model Predictive Control using accelerated constrained trajectory optimization, and (ii) Reinforcement Learning using zeroth-order optimization. We provide insights into various performance tradeoffs including sample efficiency, sim-to-real transfer, robustness to distribution shifts, and whole-body multimodality via extensive on-hardware experiments. We conclude with proposals on fusing "classical" and "learning-based" techniques for agile robot control.
APA
Abeyruwan, S., Bewley, A., Boffi, N.M., Choromanski, K.M., D’Ambrosio, D.B., Jain, D., Sanketi, P.R., Shankar, A., Sindhwani, V., Singh, S., Slotine, J. & Tu, S.. (2023). Agile Catching with Whole-Body MPC and Blackbox Policy Learning. Proceedings of The 5th Annual Learning for Dynamics and Control Conference, in Proceedings of Machine Learning Research 211:851-863 Available from https://proceedings.mlr.press/v211/abeyruwan23a.html.

Related Material