Publication

An investigation of defect inspection performance through human-robot collaboration

Conference Article

Conference

International Automatic Control Conference (CACS)

Edition

Pages

1-6

Doc link

http://dx.doi.org/10.1109/CACS67552.2025.11288169

File

Download the digital copy of the doc pdf document

Authors

Projects associated

Abstract

In modern manufacturing, visual inspection of metallic components such as solid-state drive (SSD) cases demands high precision and operational flexibility. This paper presents a collaborative human-robot inspection approach that combines high-throughput automated robot inspection with reliable and flexible human oversight. The system uses a robotic arm equipped with an end-effector-mounted 2D camera for SSD surface inspection. When robot prediction confidence exceeds a predefined threshold, the SSD is automatically classified as either Good or Defective. If confidence falls below this threshold, the system flags the SSD for human inspection. A tray with SSD cases is placed on an array of infrared proximity sensors to monitor human inspection activity. Placing an SSD back to its slot implies acceptance (Good), while removal indicates rejection (Defective). The shared plan execution is implemented using behavior trees, enabling asynchronous and intention-aware human-robot coordination. A real-time graphical interface displays the inspection state for robot-to-human communication. The approach is evaluated using different robot vision confidence thresholds, simulating more and less challenging scenarios (e.g., changes in lighting conditions). Experimental results demonstrate that under high-confidence settings, the robot alone achieved over 1600 units per hour (UPH). In low-confidence conditions requiring human intervention, the system still maintained a throughput above 900 UPH, ensuring robust defect validation. These findings confirm that the proposed collaborative solution effectively balances the repeatability and speed of robot inspection with the adaptability and reliability of human decision-making, outperforming purely human-only or robot-only baselines.



Categories

artificial intelligence, robots.

Scientific reference

S. Sanjaya, M. Ahsan, A. Olivares-Alarcos, H. Lin and G. Alenyà. An investigation of defect inspection performance through human-robot collaboration, International Automatic Control Conference, 2025, Hsinchu, Taiwan, pp. 1-6.