Monte Carlo Localization with Field Lines Observations for Simulated Humanoid Robotic Soccer

This paper presents the application of the localization technique known as Monte Carlo Localization for the estimation of the global pose of virtual robots agents in the Robocup 3D Soccer Simulation League. For this specific problem, a special formulation for this famous method is demanded in order to deal with the kidnapping problem, i. e. when the robot is moved by an external agent. We extended the classic landmark observation to incorporate also field lines observations. In this work, we will introduce the filtering algorithm and describe the stochastic modeling and implementation traits concerning this challenge. We also conducted tests in the RoboCup Soccer 3D environment to demonstrate the results obtained, especially the improvement due to the inclusion of lines observation.