Robot Trajectories Of The Neural Network-based Robust Tracking Control

In general, every complex industrial process has some repeatable properties. How to use the property of the complex system and introduce a learning mechanism to continuously accumulate knowledge from controlled plant to accomplish design and improvement of the controller on-line, is becoming a new research area of intelligent control for the complex system. In other words, all functions of on-line learning, on-line control and performance improvement of control system are integrated in an algorithm and realized by repetition of industrial processes. Thus iterative learning control is proposed as a new intelligent control methodology. It has strong advantage to solve the problem of uncertainties due to nonlinear property and external disturbance of the plant.Since neural network not only has the satisfactory capacity of approximating any nonlinear mapping but also can learn and adapt to the dynamical property of unknown system, neural network based control system has fairly strong adaptability and robustness. This paper presents a new neural network based iterative learning control algorithm, which combines iterative learning control with neural network identification for the purpose of trajectory tracking control of robot. As neural network has the ability of self-learning, that utilizes prior output data of uncertain system to estimate iteratively the static state property of system in order to achieve ideal approaching precision for identification of the positive model, a robust iterative learning control scheme on the basis of better positive model is designed. The neural network is used to identify the positive model of nonlinear system on iterative axis, which can give feed-forward action of iterative learning controller to reduce the effects of nonlinear properties and model uncertainties. Meanwhile, feedback action of iterative learning controller make joint movement follow the desired trajectory on time axis by using controlled parameters derived by the neural network. The characteristic of method is, in every process of iterative learning, after obtaining better approaching precision of network training for model identification iteratively, the feed-forward action of iterative learning control law for the next trail is constructed by output signals of the neural network, and then integrated with feedback control to track the desired trajectory of robot in real time. The feedback control is introduced to compensate effects of both errors of identification and iterative learning, so the control system can get better robustness and control precision. Simulation results indicate that the method is very effective to robotic system with unknown external disturbances, and it can also acquire satisfying tracking performance by fewer numbers of network training and iterative learning processes. While analysis performed in the paper denote that any nonlinear system with uncertain disturbances can realize high-precision tracking to any trajectory, on condition that the identification precision of neural network is good enough.