Real Time Object Detection for Competitive Robotics

Abstract

The real-time detection of objects in competitive robotics, particularly for competitions such as RoboMaster, is critical for rapid and precise decision-making. This study focuses on developing a robust object detection model utilizing YOLOv5, optimized for identifying opponent robots’ armor plates in realtime. The model was trained using publicly available RoboMaster datasets and implemented data augmentation techniques to enhance its generalization capabilities. Evaluation metrics including precision, recall, and mean Average Precision (mAP) demonstrated strong overall performance, achieving 95.1% precision, 97.2% recall, and 98.7% mAP at an IoU threshold of 50%. Despite impressive performance at moderate thresholds, stricter IoU criteria showed lower mAP scores, highlighting areas for future improvements. Ethical considerations, including privacy, transparency, and fairness, were also addressed. The advancements in this object detection model have broader implications, notably in emergency response and healthcare, signifying its potential cross-industry impact.