Abstract
RoboCup soccer middle size league requires each mobile robot accurate self-location for collision avoidance, path planning, etc. The working environment of RoboCup middle-size robots is the known soccer field, however, on the field, multiple robots are always moving and occupy much area of the field comparing robot size with field size, and sensor information changes dynamically even if the robot stands on the same position. Moreover, self-localization using images have limitation as the lighting condition changes depending on time. In this research, self-localization using image processing is introduced as the basic method, and the ball that filed has only one is employed as the landmark. The landmark information from each robot is shared by all robots and converged to improve the ball position estimation. The estimated ball position is fed back to each robot and the self-location is revised to have more accurate positions. The proposed method is implemented to RoboCup robots and evaluated through real-environment experiments.
