Graph network simulators can learn discontinuous, rigid contact dynamics

Kelsey R Allen, Tatiana Lopez Guevara, Yulia Rubanova, Kim Stachenfeld, Alvaro Sanchez-Gonzalez, Peter Battaglia, Tobias Pfaff
Proceedings of The 6th Conference on Robot Learning, PMLR 205:1157-1167, 2023.

Abstract

Recent years have seen a rise in techniques for modeling discontinuous dynamics, such as rigid contact or switching motion modes, using deep learning. A common claim is that deep networks are incapable of accurately modeling rigid-body dynamics without explicit modules for handling contacts, due to the continuous nature of how deep networks are parameterized. Here we investigate this claim with experiments on established real and simulated datasets and show that general-purpose graph network simulators, with no contact-specific assumptions, can learn and predict contact discontinuities. Furthermore, contact dynamics learned by graph network simulators capture real-world cube tossing trajectories more accurately than highly engineered robotics simulators, even when provided with only 8 – 16 trajectories. Overall, this suggests that rigid-body dynamics do not pose a fundamental challenge for deep networks with the appropriate general architecture and parameterization. Instead, our work opens new directions for considering when deep learning-based models might be preferable to traditional simulation environments for accurately modeling real-world contact dynamics.

Cite this Paper

BibTeX
@InProceedings{pmlr-v205-allen23a, title = {Graph network simulators can learn discontinuous, rigid contact dynamics}, author = {Allen, Kelsey R and Guevara, Tatiana Lopez and Rubanova, Yulia and Stachenfeld, Kim and Sanchez-Gonzalez, Alvaro and Battaglia, Peter and Pfaff, Tobias}, booktitle = {Proceedings of The 6th Conference on Robot Learning}, pages = {1157--1167}, year = {2023}, editor = {Liu, Karen and Kulic, Dana and Ichnowski, Jeff}, volume = {205}, series = {Proceedings of Machine Learning Research}, month = {14--18 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v205/allen23a/allen23a.pdf}, url = {https://proceedings.mlr.press/v205/allen23a.html}, abstract = {Recent years have seen a rise in techniques for modeling discontinuous dynamics, such as rigid contact or switching motion modes, using deep learning. A common claim is that deep networks are incapable of accurately modeling rigid-body dynamics without explicit modules for handling contacts, due to the continuous nature of how deep networks are parameterized. Here we investigate this claim with experiments on established real and simulated datasets and show that general-purpose graph network simulators, with no contact-specific assumptions, can learn and predict contact discontinuities. Furthermore, contact dynamics learned by graph network simulators capture real-world cube tossing trajectories more accurately than highly engineered robotics simulators, even when provided with only 8 – 16 trajectories. Overall, this suggests that rigid-body dynamics do not pose a fundamental challenge for deep networks with the appropriate general architecture and parameterization. Instead, our work opens new directions for considering when deep learning-based models might be preferable to traditional simulation environments for accurately modeling real-world contact dynamics.} }
Endnote
%0 Conference Paper %T Graph network simulators can learn discontinuous, rigid contact dynamics %A Kelsey R Allen %A Tatiana Lopez Guevara %A Yulia Rubanova %A Kim Stachenfeld %A Alvaro Sanchez-Gonzalez %A Peter Battaglia %A Tobias Pfaff %B Proceedings of The 6th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2023 %E Karen Liu %E Dana Kulic %E Jeff Ichnowski %F pmlr-v205-allen23a %I PMLR %P 1157--1167 %U https://proceedings.mlr.press/v205/allen23a.html %V 205 %X Recent years have seen a rise in techniques for modeling discontinuous dynamics, such as rigid contact or switching motion modes, using deep learning. A common claim is that deep networks are incapable of accurately modeling rigid-body dynamics without explicit modules for handling contacts, due to the continuous nature of how deep networks are parameterized. Here we investigate this claim with experiments on established real and simulated datasets and show that general-purpose graph network simulators, with no contact-specific assumptions, can learn and predict contact discontinuities. Furthermore, contact dynamics learned by graph network simulators capture real-world cube tossing trajectories more accurately than highly engineered robotics simulators, even when provided with only 8 – 16 trajectories. Overall, this suggests that rigid-body dynamics do not pose a fundamental challenge for deep networks with the appropriate general architecture and parameterization. Instead, our work opens new directions for considering when deep learning-based models might be preferable to traditional simulation environments for accurately modeling real-world contact dynamics.
APA
Allen, K.R., Guevara, T.L., Rubanova, Y., Stachenfeld, K., Sanchez-Gonzalez, A., Battaglia, P. & Pfaff, T.. (2023). Graph network simulators can learn discontinuous, rigid contact dynamics. Proceedings of The 6th Conference on Robot Learning, in Proceedings of Machine Learning Research 205:1157-1167 Available from https://proceedings.mlr.press/v205/allen23a.html.

Related Material