Adaptive optimal tracking control applied for a humanoid robot arm

In this paper, a recently suggested adaptive online optimal control algorithm for the infinite-horizon tracking problem of continuous-time non-linear systems with partially unknown system dynamics is modified and empirically evaluated. Since we lack complete systems knowledge a parameter identifier, which works simultaneously with the updating of the online optimal control algorithm, is introduced. We maintain tracking performance by employing an adaptive steady-state controller based on the identified system parameters and a complementary self optimizing adaptive controller, designed to stabilize the plant. To approximate the optimal value function of the Hamilton-Jacobi-Bellman equation, which is required to construct the adaptive optimal stability controller, a single layer neural network is utilized. Both the findings obtained in practice by controlling a humanoid robot-arm , as well as the results produced in simulation, demonstrate the applicability of the introduced control scheme.