Improving Gradient Computation for Differentiable Physics Simulation with Contacts

Yaofeng Desmond Zhong, Jiequn Han, Biswadip Dey, Georgia Olympia Brikis
Proceedings of The 5th Annual Learning for Dynamics and Control Conference, PMLR 211:128-141, 2023.

Abstract

Differentiable simulation enables gradients to be back-propagated through physics simulations. In this way, one can learn the dynamics and properties of a physics system by gradient-based optimization or embed the whole differentiable simulation as a layer in a deep learning model for downstream tasks, such as planning and control. However, differentiable simulation at its current stage is not perfect and might provide wrong gradients that deteriorate its performance in learning tasks. In this paper, we study differentiable rigid-body simulation with contacts. We find that existing differentiable simulation methods provide inaccurate gradients when the contact normal direction is not fixed - a general situation when the contacts are between two moving objects. We propose to improve gradient computation by continuous collision detection and leverage the time-of-impact (TOI) to calculate the post-collision velocities. We demonstrate our proposed method, referred to as TOI-Velocity, on two optimal control problems. We show that with TOI-Velocity, we are able to learn an optimal control sequence that matches the analytical solution, while without TOI-Velocity, existing differentiable simulation methods fail to do so.

Cite this Paper

BibTeX
@InProceedings{pmlr-v211-zhong23a, title = {Improving Gradient Computation for Differentiable Physics Simulation with Contacts}, author = {Zhong, Yaofeng Desmond and Han, Jiequn and Dey, Biswadip and Brikis, Georgia Olympia}, booktitle = {Proceedings of The 5th Annual Learning for Dynamics and Control Conference}, pages = {128--141}, 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/zhong23a/zhong23a.pdf}, url = {https://proceedings.mlr.press/v211/zhong23a.html}, abstract = {Differentiable simulation enables gradients to be back-propagated through physics simulations. In this way, one can learn the dynamics and properties of a physics system by gradient-based optimization or embed the whole differentiable simulation as a layer in a deep learning model for downstream tasks, such as planning and control. However, differentiable simulation at its current stage is not perfect and might provide wrong gradients that deteriorate its performance in learning tasks. In this paper, we study differentiable rigid-body simulation with contacts. We find that existing differentiable simulation methods provide inaccurate gradients when the contact normal direction is not fixed - a general situation when the contacts are between two moving objects. We propose to improve gradient computation by continuous collision detection and leverage the time-of-impact (TOI) to calculate the post-collision velocities. We demonstrate our proposed method, referred to as TOI-Velocity, on two optimal control problems. We show that with TOI-Velocity, we are able to learn an optimal control sequence that matches the analytical solution, while without TOI-Velocity, existing differentiable simulation methods fail to do so. } }
Endnote
%0 Conference Paper %T Improving Gradient Computation for Differentiable Physics Simulation with Contacts %A Yaofeng Desmond Zhong %A Jiequn Han %A Biswadip Dey %A Georgia Olympia Brikis %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-zhong23a %I PMLR %P 128--141 %U https://proceedings.mlr.press/v211/zhong23a.html %V 211 %X Differentiable simulation enables gradients to be back-propagated through physics simulations. In this way, one can learn the dynamics and properties of a physics system by gradient-based optimization or embed the whole differentiable simulation as a layer in a deep learning model for downstream tasks, such as planning and control. However, differentiable simulation at its current stage is not perfect and might provide wrong gradients that deteriorate its performance in learning tasks. In this paper, we study differentiable rigid-body simulation with contacts. We find that existing differentiable simulation methods provide inaccurate gradients when the contact normal direction is not fixed - a general situation when the contacts are between two moving objects. We propose to improve gradient computation by continuous collision detection and leverage the time-of-impact (TOI) to calculate the post-collision velocities. We demonstrate our proposed method, referred to as TOI-Velocity, on two optimal control problems. We show that with TOI-Velocity, we are able to learn an optimal control sequence that matches the analytical solution, while without TOI-Velocity, existing differentiable simulation methods fail to do so.
APA
Zhong, Y.D., Han, J., Dey, B. & Brikis, G.O.. (2023). Improving Gradient Computation for Differentiable Physics Simulation with Contacts. Proceedings of The 5th Annual Learning for Dynamics and Control Conference, in Proceedings of Machine Learning Research 211:128-141 Available from https://proceedings.mlr.press/v211/zhong23a.html.

Related Material