Simon Zimmermann

Semester Work
Supervisors: Louis Lettry and Alex Locher

Improvement of Self-Localization via Pose Correction Procedure in RoboCup

In RoboCup, humanoid NAO robots are used to perform 5-vs-5 soccer matches as a part of the Standard Platform League (SPL). In order to apply soccer tasks like kicking the ball into the opponent’s goal, the robot needs to know its position on the field. The presented work is about improving the robot’s prior pose by using Random Forest machine learning to detect the field lines of the soccer field. With these detections, the Direct Linear Transformation method is applied to find correction factors for the robot’s position. Additionally, the implementation can be used as a verification procedure to check whether the robot has lost track of its pose.