Affine Transport for Sim-to-Real Domain Adaptation

TL;DR

This paper introduces affine transport, a novel optimal transport-based method for sample-efficient sim-to-real domain adaptation in robotics, demonstrating significant error reduction in various simulation and real-world tasks.

Contribution

The paper develops affine transport with Procrustes alignment for effective domain adaptation, extending optimal transport techniques to model complex dynamics shifts.

Findings

Affine transport reduces model adaptation error significantly.

Method performs well in both simulation and real-world tasks.

Extension with Procrustes alignment captures arbitrary affine transformations.

Abstract

Sample-efficient domain adaptation is an open problem in robotics. In this paper, we present affine transport -- a variant of optimal transport, which models the mapping between state transition distributions between the source and target domains with an affine transformation. First, we derive the affine transport framework; then, we extend the basic framework with Procrustes alignment to model arbitrary affine transformations. We evaluate the method in a number of OpenAI Gym sim-to-sim experiments with simulation environments, as well as on a sim-to-real domain adaptation task of a robot hitting a hockeypuck such that it slides and stops at a target position. In each experiment, we evaluate the results when transferring between each pair of dynamics domains. The results show that affine transport can significantly reduce the model adaptation error in comparison to using the original, non-adapted dynamics model.