Real-time Neuromuscular Control of Exoskeletons

The researchers will record muscle activity in healthy participants and participants post-stroke from up to eight lower limb muscles (soleus, gastrocnemius, tibialis anterior, rectus femoris, vastus lateralis, and hamstring) during functional tasks (e.g., single-joint movement, gait, squatting, cycling). These measurements will provide the physiological dataset of lower limb movement and locomotion for the neural decoder. Then, the researchers will apply online deep learning methods for MU spiking activity decomposition from over eight muscles, and develop a real-time neural decoder. This will provide real-time decomposition of hundreds of MUs concurrently active during natural lower limb human behavior. The researchers will validate this approach by comparing our results with a gold standard, the blind source separation method. Blind source separation algorithms can separate or decompose the HD-EMG signals, a convolutive mix of MU action potentials, into the times at which individual MUs discharge their action potentials. With the decomposed MU spiking data, the researchers will develop methods to translate MU spiking activity in position, force, and hybrid commands for exoskeletons that will become a command encoder implemented into currently existing research exoskeletons that can anticipate human intent (multi-joint position and force commands) to estimate the level of assistance required by the task, (e.g., add knee torque during the stance phase). The researchers will combine the MU spiking activity decoder with the subspace projection methods into a neural real-time interface between individuals and a currently existing research lower extremity exoskeleton for locomotion augmentation. This will become an integrated high-resolution human-machine interface that can be used for real-time control of exoskeletons so that commands will be delivered at a rate higher than the muscles' electromechanical delay, i.e., the elapsed time between neural command and muscle force generation of movement. For Experiment A, the investigators will recruit healthy volunteers (n = 20) and participants post-stroke (n = 20) and complete single-joint movement and locomotor tasks to collect muscle activity data via HD-EMG. For Experiment B, the investigators will showcase the generalization of our approach recruiting and interfacing healthy volunteers (n = 20) and participants post-stroke (n = 20) with the assistive exoskeleton. Subjects will perform single-joint and locomotor tasks to calibrate the decoder, and then repeat single-joint and locomotor tasks with the decoder providing real-time assistance. Participants post-stroke will repeat up to 10 sessions to evaluate the stability of the ability of the decoder to extract motor units.