A method of single camera robocup humanoid robot localization using cooperation with walking control

The target of robocup soccer is to beat human´s world cup champion by 2050. Many kinds of technologies must be studied and developed for this purpose. In case of humanoid soccer league, the required technology level becomes higher and higher in every year. Besides, situations become closer to the practical use of humanoid robots for house, office, nursing and so on. The ideology of robocup is realized with the rules of each league. That is, there are many limitations for robocup kid-size soccer humanoid, such as robot height, foot size, and restrictions on sensors. We think that the most challenging subject under these limitations is the harmony of cost, mobility and intelligence. Our robot thus consists of single camera and small computers. In other words, our robot must estimate its field location with a costless equipment. In this paper, a method with single camera robot localization is presented. The maximum field of view is limited to 180 degrees (in 2008.) The camera can catch 2 landmarks (corner pole and goal) in usual cases. The global position information of each landmark can be used, so that the robot can get its field location by triangular surveying. The robot localization informed by two object distances is not so accurate, because the object distance from local vision system includes significant noises. Also, humanoid is not so suitable for camera carrier due to the vibrations occurred on high speed walking in actual matches. As a result, the cooperation of localization with walking control must be needed. We proposed here a method using error vector between estimation and command. In this method, the robot location can be estimated within 40 cm in most cases, with very small calculation cost.