Object Recognition in Robosoccer on Wheeled Using YOLO and ROS

Abstract

Object recognition is a critical capability for wheeled robosoccer robots operating in dynamic competition environments such as the Indonesian Wheeled Soccer Robot Contest (KRSBI-B). Limitations in real-time perception and system integration can reduce the effectiveness of autonomous navigation and opponent avoidance. This study proposes an object recognition system based on YOLOv5 integrated with the Robot Operating System (ROS) to enhance real-time perception and system responsiveness. The proposed approach employs YOLOv5 to detect opponent robots and utilizes ROS as a middleware to enable seamless communication between perception and navigation modules. Experimental results show that the system successfully detects robot objects in 11 out of 12 test scenarios, achieving an average detection confidence exceeding 0.90 within the optimal distance range of 50–350 cm. The best distance estimation performance is obtained at a distance of 350 cm, with a minimum error of 0.85%, while stable detection performance is maintained at distances up to 500 cm. These results demonstrate that the integration of YOLOv5 and ROS provides a reliable and effective solution for object recognition in wheeled robosoccer applications, supporting adaptive navigation and robust performance under dynamic operating conditions.