Democratizing Precision: High-Performance Hand Gesture Recognition via Low-Cost Exoskeleton and Physics-Informed Graph Neural Networks

Accurate hand gesture recognition is often hindered by the trade-off between the high cost of commercial data gloves and the unreliability of visual methods. To break this barrier, we introduce a low-cost 3D-printed exoskeleton glove with a total hardware cost under $20. This device accurately measures 15 degrees of freedom and achieves high-fidelity dynamic tracking, demonstrated by a Pearson correlation coefficient exceeding 0.98 against desired trajectories. We pair this hardware with a novel Angle-Graph Convolutional Network with Physical Constraints (AGCN-PC), which leverages the glove’s precise physical structure as a graph prior and incorporates inter-joint coupling constraints. On a self-collected dataset of 12 dynamic gestures from 8 subjects,it achieves 95.64% accuracy in a rigorous Leave-One-Out Cross-Validation for user generalizatio, significantly outperforming all baselines. Critically, our model exhibits remarkable data efficiency, achieving over 80% accuracy with only 33% of the training data, a level the baselines fail to reach even with the full dataset. This work validates that the synergy of affordable, high-fidelity hardware and physics-informed AI provides a powerful and accessible path toward democratizing precision gesture recognition.