Artificial Neural Network Based ILC with Application to Stroke Rehabilitation

This paper develops a model-free iterative learning control (ILC) approach that combines gradient descent and an artificial neural network (ANN) for application to a general class of nonlinear discrete-time systems. The algorithm recursively trains the ANN using all previous data collected from the system and employs a passivity condition to determine when the ANN can be used to compute the next ILC update, or whether an identification test is needed. Convergence properties are derived, as well as design choices that satisfy the passivity condition. By reducing the need to perform identification tests, the approach is shown to be significantly faster than existing model-free ILC algorithms. It is tested on a key stroke rehabilitation problem using functional electrical stimulation (FES) for hand/wrist tracking. Results using the new ILC approach show that 4 references can be tracked using only 24.8% of the experiments required by conventional ILC algorithms.