Components of Contact: An Analysis of Contact-Implicit Control Strategies on Manipulation Primitives

Grey Sarmiento, Michael Posa

Keywords: contact-implicit MPC, benchmark, manipulation

TL;DR: We present a framework for comparison and analysis of contact-implicit MPC controllers by looking at atomic tasks which isolate different contact-mode transitions.

Abstract: In uncertain, complex, and dynamic robotic tasks, contact with objects and the environments must often be planned or adapted during execution to achieve reasonable performance. Due to discontinuous and non-smooth contact dynamics, model-based control has historically struggled to find real-time solutions. However, recent years have seen the introduction of contact-implicit controllers which introduce relaxations into the dynamics and optimal control problems, forgoing optimality in favor of realizing rapid solutions for receding horizon control. Although these different methods have led to impressive, but isolated, demonstrations and results, the implications of these relaxations and resulting tradeoffs between them remain unstudied. Furthermore, evaluations lack a unified framework for more thorough comparison, discussion, and analysis. We present a method for comparing and analyzing contact-rich control through the lens of a set of "atomic tasks" which isolate the following components: make-break transitions, frictional stick-slip transitions, and contact-point repositioning. We perform a systematic comparison of three distinct contact-implicit controllers. By analyzing performance on these atomic tasks, in the context of their strategies for approximating the underlying optimal control problem, we shed light on the strengths and weaknesses of these distinct strategies. Results are analyzed in a unified simulation testbed.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 25