Path tracking and obstacles avoidance using neural networks and ultrasonic sensors for car-like robots

• Main Authors: Guo-He Huang, 黃國和
• Other Authors: Jeen-Shing Wang
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/84029016535283320624

碩士 === 國立成功大學 === 電機工程學系碩博士班 === 94 === This thesis focuses on the implementation of a car-like robot navigation hybrid control strategy characterized by global path tracking and obstacle avoidance based on neural network and ultrasonic sensors. The proposed car-like robot navigation control algorithm is capable of tracking a planned global path and avoiding the obstacle instantaneously. In order to enhance the adaptability and autonomy of the car-like robot, the algorithm adopts the concept of mode switching to provide a suitable control strategy in different situations. The control strategy consists of two modes: a tracking mode and an emergency mode. Most of the time, we set the robot in the tracking mode whose neural dynamic path-tracking algorithm enables the robot to follow a discretized planned path. In addition, we use ultrasonic sensors to detect whether there is any obstacle on the path during the navigation. Once these sensors detect the existence of the obstacles, the control strategy will switch to the emergency mode immediately. The emergency mode utilizes a neural classifier to judge the positions and possible shapes of the obstacles. Based on different positions and shapes of the objects determined by the classifier, we proceed to the obstacle avoidance control. The robot will go back to trace the global planned path after avoiding the obstacle. After software validation of the proposed algorithm, we use the ultrasonic sensors and a car-like robot developed by Parallax to verify the effectiveness and feasibility of the proposed control strategy.