Neural network augmented control for nonlinear systems

The authors describe an ongoing effort to combine a state-space controller design using feedback linearization with a neural network. They discuss an architecture for the control of a nonlinear system that augments a state-space controller with a neural network. For this architecture, they derive equations for controlling the learning rate of the neural network as a function of the tracking error in the state-space controller. The effect of noise on model learning by the neural network is examined