Deep learning for motion classification in ankle exoskeletons using surface EMG and IMU signals

Abstract Ankle exoskeletons have garnered considerable interest for their potential to enhance mobility, support rehabilitation, and reduce fall risks, particularly among the aging population. Their effectiveness depends on accurate, real-time prediction of user intentions from wearable sensor data, as even small delays or errors can compromise stability and safety. Here, we present a motion classification framework that integrates three Inertial Measurement Units (IMUs) with eight surface Electromyography (sEMG) sensors fabricated as towel-based textile electrodes, which improve comfort, durability, and usability for long-term deployment compared to traditional gel electrodes. The dataset comprises multichannel time-series recordings of five functional daily motions, enabling a realistic evaluation of exoskeleton use in everyday environments. Using this framework, Convolutional Neural Networks (CNNs) achieved an accuracy of $$99.263 \pm 0.26\%$$ 99.263 ± 0.26 % , substantially surpassing previously reported results in the field. Beyond overall accuracy, we address deployment-critical requirements: transfer learning enables reliable adaptation to new users with as few as ten calibration samples per motion, while robustness testing demonstrates that the system continues to provide stable classification even when individual sensors are disrupted. Together, these results highlight the feasibility of safe, high-accuracy, and real-world-ready exoskeleton control through deep learning combined with wearable textile electrodes and IMUs.