Pose Estimation on Soccer Robot using Data Fusion from Encoders, Inertial Sensor, and Image Data

Abstract

Middle Size League (MSL) is a division of soccer robot in the Robocup event. The research area of the robot focuses more on the ball detection and the individual capabilities of the robot and also the cooperation between one robot and the other. From the focus of the research area, the determination of the robot to the field pose is the basic ability that each robot must have to cooperate with other robots, or to improve the individual capabilities of the robot itself. One method to get a pose for a robot is the Odometry method. The main component of this method is the rotary encoder used to figure out how far the robot has moved from its initial position. However, this method is easily affected by the slip on the rotary encoder wheel, resulting in an integral error value that makes the error value bigger as the robot moves. Based on the results of experiments that has been done, in normal conditions, fusion method is more accurate than Odometry method, where the average error of fusion method is 2.07 cm at x coordinate, 2.53 cm at y coordinate, and 1.53 degrees at the robot orientation and in the Odometry method is 22.07 cm at x coordinate, 20.00 cm at y coordinate, 5.13 degrees at the robot orientation. While in the condition of a robot moving at a high speed, it turns out that the fusion method is no more accurate than the Odometry method, where the average error in the Odometry method is 17.67 cm at the x coordinate, 16.33 cm at the y coordinate, 15.5 degrees at the robot orientation and in the fusion method is 23.67 cm at x coordinate, 20.67 cm at y coordinate, 16.67 degrees at the robot orientation.