Development of a cashew nut grading system using a delta robot arm with intelligent computer vision

This is the authors' abstract. We don't add key points for this paper.

• Cost-effective Delta robot with AC servos, harmonic gears, PLC control. • Real-time YOLOv11 classifier achieving 94.2% mAP on the validation set. • Polynomial robot–camera calibration simplifying deployment and upkeep. • High-speed sorting: 1.4 s/nut (∼40/min) with success rate >98.6%. Manual grading in small and medium-scale cashew factories is error-prone, labor-intensive, and yields inconsistent quality, motivating a practical solution to automation under tight budget and hardware constraints. This study presents a cost-effective vision–robotics solution that can deliver accurate, real-time grading in such settings. The mechanical and control system comprises a Delta robot driven by AC servo motors and harmonic gearboxes with a PLC-based control system, providing precise, repeatable motion suitable for continuous industrial operation. The vision system employs a YOLOv11 deep-learning model for real-time cashew recognition and a polynomial-based camera–robot calibration that simplifies coordinate transformation without specialized camera modeling knowledge. Experimental results indicate that after training on a custom dataset, YOLOv11 model reached a mAP of 94.2% at an IoU threshold of 0.5. In experimental cashew-sorting trials, the Delta-robot system achieved an average cycle time of 1.4 s per nut (about 40 nuts/min), a sorting success ratio more than 98.6%, and sub-millimeter positioning accuracy (mean 0.4169 mm, max 1.0645 mm), demonstrating that accurate, time-efficient grading is feasible within small and medium scale factories constraints. These findings are directly relevant to cashew processors in Vietnam and similar contexts, offering a scalable route to standardized quality while reducing manual dependence.