Real-time tracking control of robot manipulators with online learning based approach

An online learning based approach to real-time tracking control of robot manipulators is proposed. This controller involves a single-layered neural network together with a traditional PD feedback loop, which inherits advantages from both the neural networks and the PD controller. It is capable of achieving real-time fine tracking control of robot manipulators under significant uncertainties, without any prior knowledge of the robot dynamics, and without any off-line training procedures. In addition, it is capable of quickly compensating sudden changes of robot dynamics. The proposed controller is computationally simple. The global system stability and convergence are proved using Lyapunov stability analysis. A model variation is presented in the discussion. The effectiveness and the efficiency are demonstrated through simulation and comparison studies.