DiffSDFSim: Differentiable Rigid-Body Dynamics With Implicit Shapes

TL;DR

This paper introduces DiffSDFSim, a differentiable physics simulation method that models complex object shapes with implicit signed distance fields, enabling shape optimization and physical parameter inference from visual data.

Contribution

It presents a novel differentiable physics framework using implicit SDFs for nonconvex shapes, supporting shape optimization and parameter inference from visual observations.

Findings

Effective shape and parameter inference from synthetic and real data

Supports nonconvex shapes in contact point calculation

Enables gradient-based shape optimization

Abstract

Differentiable physics is a powerful tool in computer vision and robotics for scene understanding and reasoning about interactions. Existing approaches have frequently been limited to objects with simple shape or shapes that are known in advance. In this paper, we propose a novel approach to differentiable physics with frictional contacts which represents object shapes implicitly using signed distance fields (SDFs). Our simulation supports contact point calculation even when the involved shapes are nonconvex. Moreover, we propose ways for differentiating the dynamics for the object shape to facilitate shape optimization using gradient-based methods. In our experiments, we demonstrate that our approach allows for model-based inference of physical parameters such as friction coefficients, mass, forces or shape parameters from trajectory and depth image observations in several challenging synthetic scenarios and a real image sequence.