Development of a Modular Bidirectional System for Wearable Sensorized Exoskeleton to Restore Sensorymotor Functions

Restoring upper limb functionality in individuals with neurological injuries remains a major challenge in rehabilitation and assistive technology. Exoskeletons have emerged as promising tools to support movement of individual or multiple joints. Recent advances in lightweight designs have enabled real-time motor assistance, yet intuitive control and sensory feedback remain limited—both crucial for dexterous motor recovery. Neurostimulation offers promising solutions to support functional restoration. In particular, Transcutaneous Electrical Nerve Stimulation (TENS) could allow to provide somatosensory feedback activating the nerve pathways non-invasively. This thesis describes the design and implementation of a modular system for controlling and sensing from a bidirectional upper-limb exoskeleton for assistive and rehabilitative applications. The motor control enables two degrees of freedom (e.g. elbow and hand), using residual surface electromyography (sEMG) signals for direct, intention-based actuation. Simultaneously, wearable pressure sensors measure fingertip forces to trigger sensory feedback via TENS. Stimulation modulates amplitude, pulse width or frequency proportionally to grasping forces and targets the median and ulnar nerves, evoking somatotopically feedback sensations. The software architecture relies on five modular classes (EMG, Classification, Sensing, Exoskeleton, and Stimulation) implemented using a parent-child class structure for reusability and flexibility. Timing management is achieved through multithreading with a main control loop running at configurable time intervals. Three data collection threads operate simultaneously during each cycle: one processes EMG signals, another record sensor’s readings and updates TENS parameters, and a third reads the exoskeleton algorithm including joint positions and force data. After data collection, another thread decodes the user intent and triggers the exoskeleton movement thread. A separate stimulation thread runs continuously, with stimulation parameters updated each cycle according to current pressure readings. The framework achieved real-time performance with a control loop under 100 ms, ensuring sufficient responsiveness for upper limb assistance. EMG classification accuracy in healthy subjects exceeded 85% for distinguishing between intended movements. Moreover, the electrical stimulation was capable of somatotopically targeting specific regions of the hand, including the palm. Participants were also able to reliably discriminate different stimulation intensities, demonstrating sensitivity to the modulation of the stimulation parameters (e.g., distinguishing between low and high amplitude). Although bidirectional control is well established in prosthetics, its integration in exoskeletons for impaired limbs remains largely unexplored. This system demonstrates independent control of hand and elbow with closed-loop sensory feedback. Its modular design supports debugging, clinical adaptation, and future scalability. Future work should improve robustness against signal variability and classification noise. Integrating pressure sensors into a sensorized glove may enhance contact realism. Moreover, hybrid control strategies combining EMG with real-time exoskeleton feedback (e.g., joint angles or torques) could offer more adaptive and natural interaction during daily activities.