Learning Control And Its Application In Robot Control

The design of the traditional control system.depends on the mathematical model of the system. Adaptive control and robust control have partly solved the uncertainty of the system model, but, in fact, due to the limitation on the system uncertainty range and the huge computation budget, there are a little application in industry. The learning control is proposed under the situation of unknown model system control. The study on the two branches of the learning control—the iterative learning control and the learning control based on neural network, is stressed to improve the intelligence of the system execution level. The detailed analysis on robot high—accuray trajectory tracking control is systematically undertaken. The main contributions of this dissertation are as follows:1. An iterative learning principle is proposed by analyzing a simple DC servomechanism system control, and, from the view point of "high—gain feedback" , the proof on the iterative control for a class of nonlinear system is also given. The experiments and simulation results on a single—degree servomechanism control and the mechanical manipulator trajectory tracking show its effectiveness. In addition, as the high—performance control of simple iterative learning method can not meet the frequently change of payload, a new iterative learning control based on the knowledge—base is presented to realize the inference computation of the control algorithm by the mathematical transform.2. The identification method based on the neural network on the separated nonlinear system is put forward explicitly using the combination of the recursive least —square estimation and backpropogation learning algorithm. The neural nets of the identified system are largely decreased by the seperated technique and the learning velocity is also improved. In addition , some improved measures on the traditional BP algorithm are proposed to strength the network learning ability.3. The learning methods on various network learning architectures are overall analyzed. The direct network controller with feedback and multi —...