Introduction: The therapeutic effectiveness of upper-limb rehabilitation robotics depend on advanced sensor integration to capture physiological intent and kinematic states. However, translating multimodal sensor modalities into a unified, adaptive control strategy remains a gap. To address this, we propose a novel multimodal sensor fusion framework designed to optimize the mapping between kinematic and physiological sensors for closed-loop robotic assistance.
Methods: The proposed framework was derived from an analysis of 20 robotic platforms, to establish optimized parameters for sensor calibration, baseline noise reduction, and algorithm design. To validate the proposed framework, we utilized the open-source Ninapro (Non-Invasive Adaptive Prosthetics) database as a standardized engineering case use. Synchronized surface electromyography (sEMG) and kinematic data from human subjects performing grasp tasks were processed through the proposed framework to evaluate intent-to-kinematic mapping efficacy.
Results: By applying the framework's fusion parameters, the system achieved an accuracy rate of 91% in predicting the correct hand grasp based on the sEMG data. By analyzing sEMG (physiological intent) with spatial joint configurations (kinematics), the framework diminished baseline noise and optimized the temporal synchronization of the sensor streams. The case-use validation on grasp classification validated the framework's algorithm mapping parameters.
Conclusions: We present a standardized architecture for multimodal sensor integration in rehabilitation robotics. By applying this theoretical model to the standardized Ninapro dataset, we demonstrate how sensor fusion solves technical heterogeneity, thereby providing a blueprint for optimizing adaptive control in upper-limb exoskeletons.