Self-localization can play important role especially when the multi autonomous robots, each equipped with distributed embedded system, cooperate as a group to achieve the desired tasks. In this study, we choose the RoboCup MSL as the target environment to investigate and develop a proper self-localization method based on the multi agent information sharing. In the proposed system, we detect the ball on the field as a common landmark for all teammate robots. Furthermore, robots can share the ball-position-information via wireless LAN and simultaneously each robot can evaluate and obtain the correct ball position based on those sharing-information. Then, each robot corrects its self-position based on the estimated landmark (ball) position. In order to evaluate the performance of the developed system, two different estimation method, probability density and weighted mean method, are implemented. This paper describes the result of experiment using the multi agent information sharing in real environment.