Morphological-Symmetry-Equivariant Heterogeneous Graph Neural Network for Robotic Dynamics Learning

TL;DR

This paper introduces MS-HGNN, a graph neural network that incorporates morphological symmetries and kinematic structures to improve robotic dynamics learning, achieving high generalizability and efficiency across diverse multi-body systems.

Contribution

The paper proposes a novel morphological-symmetry-equivariant heterogeneous graph neural network that embeds structural priors into the learning process for robotic dynamics.

Findings

Proves the morphological-symmetry-equivariant property of MS-HGNN.

Validates effectiveness on multiple quadruped robot learning tasks.

Demonstrates high generalizability and sample efficiency.

Abstract

We present a morphological-symmetry-equivariant heterogeneous graph neural network, namely MS-HGNN, for robotic dynamics learning, that integrates robotic kinematic structures and morphological symmetries into a single graph network. These structural priors are embedded into the learning architecture as constraints, ensuring high generalizability, sample and model efficiency. The proposed MS-HGNN is a versatile and general architecture that is applicable to various multi-body dynamic systems and a wide range of dynamics learning problems. We formally prove the morphological-symmetry-equivariant property of our MS-HGNN and validate its effectiveness across multiple quadruped robot learning problems using both real-world and simulated data. Our code is made publicly available at https://github.com/lunarlab-gatech/MorphSym-HGNN/.