All the introduced developments are part of the improvements carried out by the Hidalgos Team that is actively involved in the SPL. Three main goals are achieved:Firstly, the implementation of a vision-based measuring system to obtain information regarding robot surroundings. This goal requires a previous study of the camera settings and development of a new software tool for obtaining these settings and evaluating how to compensate for errors.The second goal is the definition of a local model system for modeling the sensed surrounding. Auxiliary tools must be implemented that enable the robot to deal with the information provided by the vision system.Finally, the main goal is the localization system and global modeling, which must manage the information from the local model to estimate the real position of the robot and its equivalent global model.

This article is organized as follows: in Section 2 the problem of localization and some of the most common solutions are discussed. The following section discusses the architecture used by the Hidalgos Team, while Section 4 describes the main characteristics of its perception system. The developments of the previously referred goals are contained in Section 5 (distance estimation), Section 6 (local modeling) and Section 7 (localization system). In each section, the work performed and the results obtained are carefully described. Conclusions are presented in Section 8.2.?The Localization ProblemAn increasing number of studies have focused on the localization problem and this has promoted a constant evolution of these systems.

Therefore, new localization methods have been developed as existing techniques have been improved.Early location systems were purely based on odometric readings, but this approach can provide erratic values due to the effect of foot or wheel slippage, or unpredicted slack in the joints in the case of humanoid robots. Moreover, these errors cannot be corrected because of the absence of any feedback, which could help the robot detect its errors. Thus, most sophisticated localization systems make use of sensors to provide such feedback. Sensorial information combined with an appropriate statistical procedure enables an estimation to be made of the robot position with a certain degree of accuracy. The robot soccer teams that participate GSK-3 in the SPL competition have adapted most of their localization systems.

In [1] a compilation of the methods used by some of the participant teams can be found��as well their advantages.One popular method for localization in mobile robotics is the particle filter (PF) and its derivatives. Most implementations are based on the Monte Carlo particle filter (MCL), as described in [2,3]. The MCL method represents an approximation, based on a finite number of random samples (characterized as particles) in the workspace.