Differentiable Collision Detection for a Set of Convex Primitives

TL;DR

This paper introduces DCOL, a fully differentiable collision detection framework for convex primitives, enabling gradient-based optimization in robotics applications by formulating collision detection as a convex optimization problem.

Contribution

The paper presents DCOL, a novel convex optimization-based approach for differentiable collision detection among convex primitives, allowing for gradient computation and contact analysis.

Findings

DCOL provides accurate collision metrics and contact points.

It enables gradient-based optimization in robotics tasks.

Open-source implementation available for community use.

Abstract

Collision detection between objects is critical for simulation, control, and learning for robotic systems. However, existing collision detection routines are inherently non-differentiable, limiting their applications in gradient-based optimization tools. In this work, we propose DCOL: a fast and fully differentiable collision-detection framework that reasons about collisions between a set of composable and highly expressive convex primitive shapes. This is achieved by formulating the collision detection problem as a convex optimization problem that solves for the minimum uniform scaling applied to each primitive before they intersect. The optimization problem is fully differentiable with respect to the configurations of each primitive and is able to return a collision detection metric and contact points on each object, agnostic of interpenetration. We demonstrate the capabilities of DCOL on a range of robotics problems from trajectory optimization and contact physics, and have made an open-source implementation available.