Filtro de partículas aplicado à localização de robôs móveis no domínio da Robocup Humanoide

In order to enable humanoid robots to play soccer competitively and autonomously, the robot needs to know its own position on the field, this information is important for strategy development. Informations regarding the movement of the robot and the observation made by the robot can be used to estimate its position on the domain. However, this is not a trivial task. The movements executed by the robots are inaccurate, and there are problems that can not be modeled which emerges from the physical problems of the robot. There is noise in the observations of the robot, which prevents the acquisition of precise informations regarding distance and direction, caused by the swing needed to keep the movement of the robot. Academic works about autonomous robot localization can be found for various domains, but there are a small quantity of works about the restricted domain of this work. Furthermore, the works presenting algorithms and results for this domain are not reproducible, because of the differences in hardware and software of the robots used. Thus, this work implements a localization system, based on Monte-Carlo Localization, in order to enable autonomous humanoid robots to estimate their position on the domain. The implemented system presents a method which estimates the movement of the robot over time, and methods to compute how much a particle represents the real position of the robot, besides a method to recover from position estimation errors, a method to adapt the quantity of particles in order to represent the probability distribution and a method to estimate which observation should give the best information in a near future. Experiments in simulated and real environments were performed which validates the implemented methods, and the results show that the proposed methods can effectively solve the localization problem. Finally, future works includes verifying the operation of the system in game, besides the expansion of the system to a generic domain, in order to observe the implemented methods and compare then with state of the art methods