Integrating YOLOv11 with Robotic Arms for Real-Time Shape Detection and Classification

Traditional object detection models often struggle with challenges such as variable lighting, occlusions, and the presence of multiple overlapping objects, leading to inaccuracies in shape classification. This study investigates the integration of YOLOv11 with robotic arms for real-time shape classification, focusing on enhancing the operational capabilities of autonomous systems in dynamic environments. Utilising a dataset of 4,630 annotated images representing circles, squares, and triangles, the research evaluates YOLOv11's effectiveness in accurately detecting and classifying these shapes. The model achieved a mean Average Precision (mAP) of 0.99 at an Intersection over Union (IoU) threshold of 0.50 and 0.93 at IoU 0.50–0.95, demonstrating exceptional classification accuracy. Precision and recall rates were recorded at 0.97 and 0.95, respectively, indicating robust performance in object detection tasks. The integration process involved deploying the YOLOv11 model on a 3R robotic arm, enabling real-time processing of video feeds for shape detection and sorting. Detection times averaged around 0.8 seconds, while sorting times ranged from 2.6 to 5.1 seconds, showcasing the system's efficiency in operational settings. The study highlights the critical relationship between confidence thresholds and prediction accuracy, emphasising the need for careful calibration to mitigate misclassifications. Overall, the findings affirm YOLOv11's potential as a powerful tool for real-time object classification in robotics, suggesting avenues for future research to optimise performance and expand application domains in autonomous systems. Keywords: Autonomous, Confidence threshold, Dataset, Precision, Recall, Sorting