The Trajectory Tracking Of Mobile Robot Based On Neural Network

Nonholonomic mobile robot has been an important subject of Automation area in research and development and trajectory tracking is an import technology of mobile robot. Therefore the investigation of mobile robot has a significant theoretical and practical meaning. Based on the reading and summarizing a lot of literatures references, this thesis studies the control problems focusing on trajectory tracking, constructing trajectory tracking controller with improved BP neural network and fuzzy neural network.Firstly improve BP neural netwok and make application to trajectory tracking of mobile robot. Aiming at the speed of BP neural network is slow and cannot meet the real-time request in complex multiple input conditions, a method of partitioning it into several smaller subnets in order to obtain more efficient computation is proposed. And using the idea of the improved BP neural network study the trajectory tracking problem of mobile robot with unknown parameters and interferences. A improved computed-torque control by the integration of a kinematic controller and a dynamic controller is proposed. The kinematic controller is designed by the backstepping technology and the dynamic controller is optimized by the improved BP neural network.Then aiming at the problem of neural network that can not use priori knowledge and has not the ability to reason, fuzzy neural network is applied to design the trajectory tracking controller of mobile robot. Use posture errors of mobile robot to train the fuzzy neural network and then obtain linear velocity and angular velocity to control mobile robot tracking reference trajectory. And also put a date supervision link in front of the fuzzy neural network in order to filter mutable data for increasing the stability of the system.At last, use MATLAB software simulate and compare mobile robot tracking reference trajectory based on the two algorithms of the improved BP neural network and the fuzzy neural network respectively to validate the correctness and feasibility of the controller design. And the fuzzy neural network controller is much better than the improved neural network controller for more accurate trajectory tracking and quicker converging speed.