Wearable neural interface for decoding ankle dynamics powered by automatic muscle localization using a sensorized garment
Year: 2026
Type: preprint
Abstract: Decoding neural drive during locomotion outside the laboratory requires wearable interfaces capable of reliably recording muscle activity without expert electrode placement. Here, we present a proof-of-concept wearable neural interface that integrates a 64-electrode leg garment, anatomical model-based muscle localization, and motor unit decomposition to decode neural activation and estimate ankle joint torque during walking. In nine healthy individuals walking on a treadmill with garment rotations of up to ±20°, the proposed algorithm successfully localized the plantar-flexors, with high similarity to manually selected reference electrodes (R² ≥ 0.9), and improved dorsiflexor localization compared with a heuristic-based approach. Importantly, motor units could be identified directly from the garment electrodes across all localized muscles. Motor unit-derived neural drives provided an alternative representation of muscle activation to conventional EMG envelopes and, when used to drive calibrated musculoskeletal models, enabled ankle torque estimates with comparable accuracy (R² of 0.68 versus 0.63 for EMG-derived activations). These findings demonstrate the feasibility of integrating motor unit identification into wearable sensorized garments, even with practical electrode configurations and without expert placement. This provides an important step towards wearable neural interfaces for monitoring neuromuscular function and estimating joint biomechanics beyond conventional laboratory settings
Institution: University of Twente
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7235837816
