Predicting Real Distance for Wheeled Soccer Robot using YOLO Architecture

Abstract

This study presents a real-distance estimation system for wheeled soccer robots that integrates stereo vision cameras with a YOLO-based object detection algorithm to support accurate perception for game strategies. Experimental evaluations using 10 ball samples and 6 robot scenarios demonstrate that the proposed system achieves an average distance estimation accuracy of 96.7% with a low error margin of ±2.3% and a confidence level of 99%, indicating high reliability in object detection and metric distance measurement. The results confirm that stereo vision combined with deep learning provides precise spatial information suitable for dynamic soccer robot environments, enabling improved positioning and decision-making. While the system performs robustly under standard conditions, future work will address performance degradation caused by lighting variations, explore newer YOLO model architectures, and incorporate artificial intelligence–based adaptive strategies to further enhance autonomy and competitiveness in wheeled soccer robot applications.