Real-time fine motion control of robot manipulators with unknown dynamics

A novel neural network based approach is proposed for real-time fine motion control of robot manipulators without any knowledge of the robot dynamics and subject to significant dynamics uncertainties. The controller structure consists of a simple feedforward neural network and a PD feedback loop, which inherits advantages from both the neural network based controllers and the traditional PD-type controllers. By taking advantage of the robot regressor dynamics, the neural network assumes a single-layer structure, and the learning algorithm is computationally efficient. The real-time fine motion control of robot manipulators is achieved through the on-line learning of the neural network without any off-line training procedures. The PD control loop guarantees the global stability during the learning period of the neural network. In addition, the proposed controller does not require any knowledge of the robot dynamics and is capable of quickly compensating sudden changes in the robot dynamics. The global system stability and convergence are proved using a Lyapunov stability theory. The proposed controller is applied to track an elliptic trajectory and to compensate a sudden change in the robot dynamics in real-time. The effectiveness and the efficiency of the proposed controller are demonstrated through simulation and comparison studies.