Neural network-based adaptive control of a class of uncertain nonlinear systems (Q1369415)

Consider the system: \[ \begin{aligned} \dot x & =f(x) +\delta f(x)+ g(x)u \\ y & =h(x), \end{aligned} \] where \(\delta f(x)\) is unknown and \((f(x),g(x))\) are a feedback linearizable pair. To deal with the uncertainty, this paper proposes a sliding mode scheme to augment the typical feedback linearization control law for tracking. The augmented control law includes an estimate of a form of the uncertainty terms. This estimate is made using a neural network (Gaussian radial basis network). A theorem is proven that provides a global convergence result.