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

2024-12-02

arXiv.org

AI Summary

This work addresses poor generalization and low sample/model efficiency in robot dynamics learning. Methodologically, we propose a heterogeneous graph neural network (HGNN) that integrates kinematic structure and morphological symmetry. We formally define and implement morphological symmetry equivariance for the first time, unifying rigid-body kinematics, Lie-group symmetries, and multibody system topology into a heterogeneous graph representation, while embedding geometric and physical priors via equivariance constraints. Theoretically, we prove the validity of the proposed equivariance for multibody systems. Experimentally, our method achieves significant improvements across multiple quadruped robot simulation and real-world hardware platforms: average prediction accuracy increases by +12.7%, and data efficiency improves markedly—requiring 40% less training data to attain equivalent performance. All code is publicly released.

Technology Category

• Intelligent Robots: Learning & Optimization for ROB
• Machine Learning: Neuro-Symbolic Learning
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

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/.

Problem

• Learning robotic dynamics using symmetry-equivariant graph networks
• Integrating kinematic structures and symmetries for efficiency
• Generalizable architecture for multi-body dynamic systems

Innovation

• Morphological-symmetry-equivariant heterogeneous graph neural network
• Integrates kinematic structures and symmetries into graph
• Ensures high generalizability and model efficiency