gradSim: Differentiable simulation for system identification and visuomotor control

TL;DR

gradSim introduces a differentiable simulation framework that jointly models scene dynamics and image formation, enabling physical property estimation and visuomotor control directly from videos without requiring 3D labels.

Contribution

It combines differentiable multiphysics simulation and rendering to enable end-to-end learning of physical properties from videos without 3D supervision.

Findings

Achieves competitive or superior performance to 3D-supervised methods.

Enables learning physical properties for deformable objects and cloth.

Supports visuomotor control tasks without state-based supervision.

Abstract

We consider the problem of estimating an object's physical properties such as mass, friction, and elasticity directly from video sequences. Such a system identification problem is fundamentally ill-posed due to the loss of information during image formation. Current solutions require precise 3D labels which are labor-intensive to gather, and infeasible to create for many systems such as deformable solids or cloth. We present gradSim, a framework that overcomes the dependence on 3D supervision by leveraging differentiable multiphysics simulation and differentiable rendering to jointly model the evolution of scene dynamics and image formation. This novel combination enables backpropagation from pixels in a video sequence through to the underlying physical attributes that generated them. Moreover, our unified computation graph -- spanning from the dynamics and through the rendering process -- enables learning in challenging visuomotor control tasks, without relying on state-based (3D) supervision, while obtaining performance competitive to or better than techniques that rely on precise 3D labels.