Automated tuning of a closed-loop hand grasp neuroprosthesis

An automated tuning algorithm was developed to reduce the time and skill required to tune a closed-loop hand grasp neuroprosthesis. The time reduction results from simultaneous tuning of four gain parameters controlling the dynamic response of the system, and from automation of the calculation and decision processes. The new tuning method is therefore an automated parallel tuning method, replacing a manual sequential method in which only one parameter at a time was tuned. RMS error between the step input and the grasp output is minimized, with absence of oscillation as a constraint. The difference between the system's RMS ramp tracking errors for the two tuning methods was less than 1% of the ramp size regardless of the initial values of the parameters, implying that the tuning methods were equivalent. However, the parallel tuning method was faster and required fewer trials than the sequential method. The capability of the closed-loop system to regulate grasp output in the presence of disturbances was compared with the capability without feedback. Patients were instructed to either grasp an object at a certain force level or to match a certain grasp opening. They would then lock their command at a fixed value, and either remain immobile to test time dependence or pronate and supinate their forearm to test postural disturbances. With closed-loop control, the grasp output was better regulated in the presence of disturbances, with an average output variance 60% lower than without feedback control.