The Design of Real-Time Obstacle-Avoidance Strategy for the Formation Control of a Multi-Agent System

• Main Authors: Kuang-Yu Wu, 吳光俞
• Other Authors: Li-Sheng Wang
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/94031303328325831914

碩士 === 國立臺灣大學 === 應用力學研究所 === 102 === The purpose of this thesis is to integrate the hardware and the path-planning algorithm, and then to design a real-time obstacle-avoidance strategy for a multi-vehicle system, which is realized with a system of three vehicles. The pseudo-rigid formation design algorithm was adopted to design the formation of the multi-vehicle system. The algorithm applies the idea of pseudo-rigid body theory to the formation design, which is determined by a homogenous deformation tensor such that stretch, rotation, and shear are allowed. Comparing to rigid body formation design, pseudo-rigid formation has a better adaptability to environments of higher complexity. In this method, the Rapidly-Exploring Random Tree (RRT) method was first used along with the techniques of route adjustment to obtain the route of the formation center. The deformation matrix is then found by the method of virtual potential function, from which the route of each vehicle is computed. We utilize Microsoft’s Kinect to initially detect environmental objects and execute path-planning design first, and then the function of real-time obstacle-avoidance during the motion of vehicles is implemented. With regard to the problem of obstacle-avoidance, we discuss the strategy for the single vehicle first, which is followed by the discussion for the case of a three-vehicle system. Fuzzy theory and the coordinated control algorithm were used for the multi-vehicle system to adjust the parameter of the deformation matrix in real-time. The vehicles were driven to avoid the obstacles in short distance, while the pseudo-rigid formation is kept during the process. Experimental results show that the strategy proposed in this thesis for a multi-vehicle system is feasible.