ROBUST COMPONENT DETECTION FOR FLEXIBLE MANUFACTURING: A DEEP LEARNING APPROACH TO TRAY-FREE OBJECT RECOGNITION UNDER VARIABLE LIGHTING

ICLR 2026 Conference Submission16796 Authors

Keywords: Industry 4.0; computer vision; Mask R-CNN; object detection; smart manufacturing; variable lighting; industrial vision; robotics

Abstract:

Flexible manufacturing systems in Industry 4.0 require robots that can handle
objects in unstructured environments without rigid positioning constraints. This
paper presents a computer vision system that enables industrial robots to detect and
pick up pen components in arbitrary orientations without the need for structured
trays, while maintaining robust performance under varying lighting conditions.
We implement and evaluate a Mask R-CNN-based approach in a complete pen
production line, addressing three key challenges: object recognition without positional constraints, robustness to extreme lighting changes, and reliable performance
with cost-effective cameras. Our system achieves 95% recognition accuracy under
diverse lighting conditions and eliminates the need for structured component placement, resulting in significant improvements in manufacturing flexibility and overall
robustness. This approach has been validated through extensive experiments under
four distinct lighting scenarios. These results demonstrate its practical applicability
for real-world industrial deployment.

Primary Area: applications to robotics, autonomy, planning

Submission Number: 16796