A Robot Vision Algorithm For Navigating In And Creating A Topological Map Of A Reconfigurable Maze

In this paper we present a neural network approach to solving the problem of a robot agent (Mazebot) navigating in and creating a topological map of a reconfigurable maze. The robotics system used is based on an SRV-1 Robot extended both in hardware and software to accomplish the task. The main algorithm of the system is vision based, requiring only a single camera and a dead reckoning sensor. For the purposes of our algorithm a database of images from various maze configurations has been created. Neural Networks are utilized to train the agent at first and later to analyze features extracted from the images and enable agent navigation inside the maze. The advantage of our approach lies in the minimal number of sensors required by the robot agent to achieve success in its task.