Supervisors: Louis Lettry and Alex Locher
Self-localization is a key aspect of navigation and decision making for autonomous robots. In RoboCup, where humanoid NAO robots perform soccer matches against each other, the knowledge of the robot's own position is critical for gameplay. Therefore, a fast and reliable relocalization method is required in case the robots lose track of their position during the game. The goal of this work is to find the position of the robot on the field based solely on the detected field lines. First, the previously established global localization method is investigated and improved. It uses the detection of pre-defined shapes on the field (such as T-junctions) to create hypotheses of the current robot position. However, this method is not stable against errors in line detection or when a crucial part of the line is not detected. Therefore, a second self-localization method is established, using a point-to-line iterative closest point method, and fusing this information with the robot's particle filter.