Contact Points Discovery for Soft-Body Manipulations with Differentiable Physics

Abstract

Differentiable physics solver often gets stuck when the initial contact points of the end effectors are sub-optimal or when performing multi-stage tasks that require contact point switching, which often leads to many local minima. To address this challenge, we propose a contact point discovery approach (CPDeform) that guides the stand-alone differentiable physics solver to deform various soft-body plasticines.

Publication
ICLR, 2022 (Spotlight, 5%)

Dataset Illustration

Dataset Image

Airplane

Chair

Writer

Move++

Rope++

Code Release

Please check out our GitHub repository CPDeform.

Citation

@InProceedings{li2022contact,
author = {Li, Sizhe and Huang, Zhiao and Du, Tao and Su, Hao and Tenenbaum, Joshua and Gan, Chuang},
title = {{C}ontact {P}oints {D}iscovery for {S}oft-{B}ody {M}anipulations with {D}ifferentiable {P}hysics},
booktitle = {International Conference on Learning Representations (ICLR)},
year = {2022}}

Acknowledgement

We thank Hannah Skye Dunnigan for her help on graphic design. This work was supported by MIT-IBM Watson AI Lab and its member company Nexplore, ONR MURI (N00014-13-1-0333), DARPA Machine Common Sense program, ONR (N00014-18-1-2847) and MERL.

Sizhe Lester Li
Sizhe Lester Li
李思哲

My research interests span robot learning, vision, and physics simulation. Currently, I develop methods for robots to learn to interact with deformable objects with challenging dynamics.