Exciting Contact Modes in Differentiable Simulations for Robot Learning

TL;DR

This paper proposes an optimal experimental design approach derived from information-theoretic methods to identify and search for information-rich contact modes through the use of contact-implicit optimization, and demonstrates the approach on a robot parameter estimation problem with unknown inertial and kinematic parameters.

Abstract

In this paper, we explore an approach to actively plan and excite contact modes in differentiable simulators as a means to tighten the sim-to-real gap. We propose an optimal experimental design approach derived from information-theoretic methods to identify and search for information-rich contact modes through the use of contact-implicit optimization. We demonstrate our approach on a robot parameter estimation problem with unknown inertial and kinematic parameters which actively seeks contacts with a nearby surface. We show that our approach improves the identification of unknown parameter estimates over experimental runs by an estimate error reduction of at least $\sim 84\%$ when compared to a random sampling baseline, with significantly higher information gains.

Figures (2)

• Figure 1: Results for the planar block throwing and three-link planar arm evaluated over 20 experiments. Compared to uniform random sampling methods, we observe $\sim 97 \%$ and $\sim 84 \%$ parameter estimates error reduction for the (a) block and (c) arm experiments, respectively, and we observe higher information gains over experiments for both the (b) block and (d) arm.
• Figure 2: Arm (a) and block throwing (b) example contact-aware experiment via our proposed approach.