Deep Neural Network Real-Time Control of a Motorized Functional Electrical Stimulation Cycle With an Uncertain Time-Varying Electromechanical Delay

Abstract

Closed-loop functional electrical stimulation (FES) control methods are developed to facilitate motor-assisted cycling as a rehabilitative strategy for individuals with neurological disorders. One challenge for this type of control design is accounting for an input delay called the electromechanical delay (EMD) that exists between stimulation and the resultant muscle force. The EMD can cause an otherwise stable system to become unstable. A real-time deep neural network (DNN)-based motor control architecture is used to estimate the nonlinear and uncertain dynamics of each leg of the cycle-rider system. The DNN estimate of the system’s dynamics updates in real-time and is used as a feedforward term in the motor controller allowing the cycle crank to meet position and cadence tracking objectives. The nonsmooth Lyapunov-based stability analysis proves semiglobal asymptotic tracking.