Humanoid robot localization in a soccer field using Deep Learning

Abstract

With the evolution of humanoid robotics and its increasing use in diverse environments and tasks, it is imperative that the robot can interact with the environment and, therefore, understand it accurately to execute decision making. In this work, we presented the process of collecting a new dataset for simulated soccer scenes. We simulated the RoboCup Humanoid challenge and collected over 200k images that contain up to 4 classes of objects, depth estimation, and bounding boxes. We then trained a modified multiclass version of J-MOD2 to validate the dataset and provide the landmarks distances to a Monte Carlo localization algorithm in order to estimate the robot position on the field.