Dataset and Baseline Experiments for Self-Localization and Tracking in the RoboCup Small Size League

Abstract

The RoboCup Small Size League (SSL) and the Vision Blackout challenge, aim to increase robot autonomy and encourage teams to explore local sensing and processing Thus, this paper proposes the use of an open-source dataset for the self-localization and tracking of SSL robots, which was recorded onboard and contains ground-truth poses, estimated by ssl-vision, odometry and speed data, computed onboard. Recordings contain nine different scenarios, with multiple paths and velocities, and goals’ detections were annotated. A Monte Carlo Localization (MCL) algorithm implementation is described and tested on the proposed dataset, and evaluation metrics and experiment results from the self-localization and odometry algorithms using the collected data are presented. This work aims to enable SSL teams to explore and evaluate solutions for robot self-localization and tracking problems, providing baseline experiments and metrics for objective comparisons of future solutions.