A Study on Robotic Arm Grasping for Rail Transit Maintenance Scene Based on Improved GGCNN

In rail transit maintenance scenarios, robotic arm grasping tasks need to cope with complex environments (metal reflections, oil contamination), diverse targets (bolts, cables, insulators), and high safety requirements. GGCNN (Generative Grasping Convolutional Neural Network) excels in grasping tasks with its lightweight and real-time performance but is not sufficiently powerful for small target detection, anti-interference and dynamic scene adaptation are insufficient. In this paper, an improved GGCNN framework is proposed to significantly enhance the grasping performance in rail traffic scenarios by means of anti-jamming module, high-resolution feature extraction, lightweight design and simulation-to-reality migration. The experiments validate the superiority of the improved model on a self-built rail traffic dataset and a real robotic arm, with a 92.3% grasping success rate, a 20% improvement in anti-jamming, and a real-time performance that meets the requirements of embedded deployment. This study provides an efficient solution for rail transit automated maintenance and lays the foundation for intelligent inspection robot application.