Vision-based monte-carlo localization for humanoid soccer robots

Abstract

In order to solve the self-localization problem, the Monte-Carlo Localization is proposed as a technique which is able to solve any localization problem. However, the implemented algorithm needs to be adapted to the robot and domain. Thus, this work presents a novel implementation for humanoid robots whose main sensory input is a camera, to be used in the domain of RoboCup Humanoid Soccer League. The paper proposes motion and observation models designed for the domain, and a method to determine the quantity of particles needed to represent the probability distribution. Finally, the proposals are validated by simulated experiments.