Visual navigation system for mobile robots

Abstract

This dissertation work is devoted to the study on visual navigation system for mobile robots. For this, four topics are presented as followings:First, a range sensor data based path planner is proposed for multiple robots with obstacles environment. Using Hopfield neural optimization network, an energy function is defined in terms of the range sensor data and position information for each robot, and then, a collision-free path is planned through minimization of the energy function. The proposed algorithm can be sued for a real-time path planner for mobile robots, and is found to be appropirate for autonomous vehicle since it is performed in a totally localized fashion. Second, a 3D self-positioning algorithm is proposed. From the homogeneous transformations between the robot coordinate frames, we derive the minimal number of guide points required to determine the position and orientation of a mobile robot. And, a closed form self-positioning algorithm is proposed using a streo camera system and a simple guide mark. The proposed algorithm is advantageous of its 3D capability, and is practically useful with its computationally effective closed form. Third, a fast 3D depth map estimation algorithm is proposed. In this work, the flows of stereo image signals are dynamically controlled to satisfy the matching condition of their intenstity functionals. Then, the disparity and its 3D depth are estimated from these controls for the image signals, sequentially. The proposed method is applicable as a real time depth estimator with low cost hardware for robotic applications. Finally, we proposed a design approach to the efficient CMAC system for servo control. We analyzed the CMAC computational scheme, and found the effective CMAC memory size which is feasible in most practical applications. Using the proposed algorithm, the CMAC system can be implemented with a small size of memory and its new address mapping method, and the learning efficiency of the CMAC is improved.