Affine Transport for Sim-to-Real Domain Adaptation

2021/05/25 by Anton Mallasto, Mallasto, Anton, Karol Arndt +7
Physics and Astronomy · Computer Science · #Model Reduction and Neural Networks #Advanced Vision and Imaging #Domain Adaptation and Few-Shot Learning

paper · pdf · doi:10.48550/arxiv.2105.11739

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.

Citations

• The Fréchet distance between multivariate normal distributions
• Polar factorization and monotone rearrangement of vector‐valued functions
• Large-Scale Optimal Transport and Mapping Estimation

Related

• Generalizable Domain Adaptation for Sim-and-Real Policy Co-Training
• Understanding Domain Randomization for Sim-to-real Transfer
• Policy Transfer via Kinematic Domain Randomization and Adaptation
• Meta Reinforcement Learning for Sim-to-real Domain Adaptation
• Predicting Sim-to-Real Transfer with Probabilistic Dynamics Models