Localization of Wheeled Soccer Robots Using Particle Filter Algorithm

Localization is a method used to determine the position of an object. In mobile robots, especially wheeled soccer robots, determining the location of robots against landmarks is needed. It can determine the movements and strategies of soccer robots in a game. There are several methods can be used for localization, namely the odometry method and the extended Kalman filter. But the methods are strongly influenced by the results of reading sensors and tire slips on the robots. In this article, we proposed a particle filter algorithm for localization of wheeled soccer robots. This particle filter method is suitable for non-linear, non-Gaussian systems, and can be used in a real-time environment. We applied the particle filter algorithm to localize the position of the wheeled soccer robot. The particle filter algorithm generally has three basic stages, prediction stage, weight update and resampling. As initialization process, the particles are spread randomly or around the initial position of the robot. The prediction stage is done during the robot movement using a motion model applied to each particle. After that, the weight of each particle is updated based on the likelihood of particle data to the robot data. The resampling step is performed to update the weight of each particle. Particles with the smallest weight value will disappear by themselves and particles with large weight will retain. The position presumption is obtained through the estimation process that is by averaging the position of all particles. The experiment is done using simulation on the field with 450 × 300 pixels or 900 × 600 cm which each pixel represent 2 cm on the real soccer for competition. The particles used in the experiment are 100, 200, 400, 800, and 1600 particles. The experiment results show the satisfied result with number of particles error is 100 particles and the error is between 0.363 cm and 23.055 cm.