A Vision-driven Robotic Dynamic Grasping System for Anomaly Objects in Dynamic Production Scenarios

In 2021, global manufacturing activity had rebounded to pre-pandemic levels, accounting for approximately 17% of worldwide gross domestic product on average, and the EU contributing 14.9%, solidifying its position as the region’s most economically significant industrial sector . However, the implement of advanced robotic systems has shown an annual global increment of 12% with the most active being in the industrial sector , which is expected to reach USD 60,562 million by 2030 . To sustain its position and enhance its manufacturing competitiveness, EU manufacturing industries must advance Green Deal-driven sustainable industrial growth , accelerate smart automation under Industry 5.0, and enable circular economy transitions , while upholding stringent product quality and safety standards. Thus, developing advanced, production-challenge-tailored robotic systems has grown ever more critical. In current industrial settings, vision-driven robotic grasping models are predominantly deployed for picking and sorting tasks on production lines , whereas defective products are typically identified by vision systems but still often removed by human laborers. Academically, research in anomaly detection (AD) and grasping pose detection (GPD) has been progressing independently due to the limited industrial datasets, fewer defect types and temporal dependencies. Against this backdrop, the Vision-driven Robotic Defective Grasping System (VRDGS) emerges as a solution designed to address these challenges head-on. The robotic griper is required to quickly identify anomaly objects and remove them without downtime, ensuring production efficiency and quality.