🤖 AI Summary
This work addresses three key challenges in contact sensing for legged robots: poor generalization, low sample efficiency, and model redundancy. To this end, we propose Morphology-Informed Heterogeneous Graph Neural Networks (MI-HGNN). Methodologically, we explicitly represent the robot’s rigid-body topology—as a heterogeneous graph where joints serve as nodes and links as edges—and incorporate morphological constraints and dynamics priors to achieve seamless integration of model-based knowledge and data-driven learning. Compared to state-of-the-art methods relying on morphological symmetry, MI-HGNN achieves an 8.4% improvement in accuracy while reducing parameter count to just 0.21%. Trained jointly on multi-source simulation and real-robot data, the method significantly enhances robustness and cross-scenario generalization. Experimental validation on a physical quadruped platform confirms its effectiveness and practical deployability.
📝 Abstract
We present a Morphology-Informed Heterogeneous Graph Neural Network (MI-HGNN) for learning-based contact perception. The architecture and connectivity of the MI-HGNN are constructed from the robot morphology, in which nodes and edges are robot joints and links, respectively. By incorporating the morphology-informed constraints into a neural network, we improve a learning-based approach using model-based knowledge. We apply the proposed MI-HGNN to two contact perception problems, and conduct extensive experiments using both real-world and simulated data collected using two quadruped robots. Our experiments demonstrate the superiority of our method in terms of effectiveness, generalization ability, model efficiency, and sample efficiency. Our MI-HGNN improved the performance of a state-of-the-art model that leverages robot morphological symmetry by 8.4% with only 0.21% of its parameters. Although MI-HGNN is applied to contact perception problems for legged robots in this work, it can be seamlessly applied to other types of multi-body dynamical systems and has the potential to improve other robot learning frameworks. Our code is made publicly available at https://github.com/lunarlab-gatech/Morphology-Informed-HGNN.