Comparison of contact models for robotics simulation

Abstract

Models of rigid-body dynamics are frequently used in robotics. Smooth dynamics, even with joint constraints, can be simulated efficiently and accurately. However, some of the most difficult and important problems in robotics involve contacts between bodies, which introduce non-smoothness. Various approaches have been proposed to simulate contact dynamics. Some relax the problem to improve computational properties. However, the tradeoffs and differences between these models are not well characterized.

This thesis aims to identify and explain differences in a case study of four contact models. Simulated trajectories from the models are compared in seven different scenarios, which include wedging a box, grasping a sphere, and dropping spheres into a box. The models exhibit qualitatively different behaviors in the scenarios, some of which only become evident with multiple simultaneous contacts. The results can help inform roboticists when choosing contact models for specific applications or isolating sources of unphysical behavior.