Robust Tracking of Multiple Soccer Robots Using Random Finite Sets

Having a good estimation of the robot-players positions is becoming imperative to accomplish high level tasks in any RoboCup League. Classical approaches use a vector representation of the robot positions and Bayesian filters to propagate them over time. However, these approaches have data association problems in real game situations. In order to tackle this issue, this paper presents a new method for building robot maps using Random Finite Sets (RFS). The method is applied to the problem of estimating the position of the teammates and opponents in the SPL league. Considering the computational capabilities of Nao robots, the GM-PHD implementation of RFS is used. In this implementation, the estimations of the robot positions and the robot observations are represented using Mixture of Gaussians, but instead of associating a robot or an observation to a given Gaussian, the weight of each Gaussian maintains an estimation of the number of robots that it represents. The proposed method is validated in several real game situations and compared with a classical EKF based approach. The proposed GM-PHD method shows a much better performance, being able to deal with most of the data association problems, even being able to manage complex situations such as robot kidnappings.