Lightweight YOLO Object Detection Algorithms for Vision-Based Robotic Arms: A Review

Abstract

To address the requirements of vision-based robotic manipulators for object detection accuracy, real-time performance, and lightweight deployment, this paper reviews lightweight You Only Look Once (YOLO) object detection techniques and their applications in robotic vision. First, the main techniques, including structural lightweighting, model compression, and edge deployment, are summarized. Representative algorithms for vision-based robotic manipulators are then categorized into three groups: backbone lightweighting, local structural optimization, and model compression, with their detection performance and lightweighting characteristics analyzed. On this basis, the limitations of existing methods are discussed in terms of the trade-off between accuracy and efficiency, cross-scenario generalization, coordination between visual perception and grasp execution, and comprehensive evaluation. Finally, future research directions are outlined, including multi-strategy collaborative optimization, cross-domain adaptation, and system-level evaluation for robotic manipulators.