Self-Localization System for Robots Using Random Dot Floor Patterns

Various types of service robots have recently been developed for guarding facilities, caring for the elderly, carrying objects, and cleaning buildings. As barrier-free facilities improve and their use expands, these robots have more space within which to move inside buildings. Yet robots that move autonomously rely on position-detection systems. Though improving rapidly, these systems are far from perfect in determining positions in certain situations, especially when robots navigate large areas or cross various locations. Our group is working to solve this problem by developing a position-detection system using random-dot patterns on a floor. First, we construct a floor with a random-dot pattern and register the positions of all of the dots into a database. As the robot moves across the floor, a camera on the robot captures an image of the floor beneath it and crops the dot pattern in the image. The cropped dot pattern is matched to the dot patterns in the database to determine the position of the robot and the direction in which the robot is facing or moving. In this paper we propose a self-localization system and matching algorithms derived from a space technology and present the results of several experiments.