Answers are grounded in this paper and cite their sources.
Advanced biomechanical models are required to support the operations of exoskeleton systems employed for patient rehabilitation. However, current models are unable to account adequately for the non-ideal attachment effects associated with these exoskeleton systems that cannot be fully represented by an ideal rigid-connection model. Therefore, these attachment effects represent sources of residual dynamics remaining unaccounted for by biomechanical models that may jeopardize safe operations. The present work addresses this issue by proposing a physics-informed machine learning approach driven by model–data fusion for identifying the dynamic parameters of coupled lower-limb–exoskeleton systems. First, a baseline coupled dynamic-system model is established with explicit physical meaning under the assumption that the lower limb and exoskeleton are rigidly connected, and the model is linearly parameterized. Then, a data-driven residual compensation method based on a backpropagation network is introduced to capture un-modeled dynamics induced by non-ideal system interfaces. The results of experiments conducted using an anthropomorphic lower-limb dummy coupled with a rehabilitation exoskeleton demonstrate that the proposed residual compensation framework substantially improves joint-torque prediction over that of the physics-based model alone, while preserving model interpretability. The results of momentum-observer-based collision detection experiments further demonstrate that the proposed approach can support model-dependent safety tasks in wearable exoskeleton systems.
No takes yet. Share an insight, caveat, or question.
Fang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: