隨著AI技術的進步，智慧化機器人技術已然成為趨勢，它能夠擁有如同人類般的感知能力，以便執行更加複雜的任務與目標，電子廠內，對於電子零件所導入的自動化產線已相當成熟，但對於少量多樣的電子連結器裝配任務，建置其自動化產線與振動盤相對耗費成本，故本研究旨於打造一套使用機械手臂、機器視覺及深度學習的散料夾取系統，以適應微小電子連結器的複雜特徵與環境，將其嵌入本系開發之類產線架構中，成功實現了彈性產線，解決自動化產線與振動盤的高昂成本及缺工問題。
本論文使用YOLOv8物件偵測模型，並以資料擴增(Data augmentation)的方法，以當前環境亮度±25%之光源進行訓練，進而提升穩健性，物件偵測模型會自動產生許多的候選框(Boundingbox)，透過Non Maximum Suppression(NMS)方法，濾除多餘的候選框，為了提升組裝成功率，本研究設計了用於整理微小電子連結器姿態的轉接治具，根據物件框選中心與影像質心位置計算其夾取之開口方向，由於組裝插件有方向性需求，故需以開口向上、下以及正反面，四種特徵設定其放料姿態，置放於轉接治具中，接著會等待轉盤的訊號收發，將轉接治具上的電子連結器夾持至插端子治具上，為了提升組裝準確率，亦開發了一套標定演算法與標定治具，嵌入於插入微小電子連結器的治具上，以判定組裝的插槽實際位置，完成後，透過IO訊號發送任務結束訊號給轉盤，以切換下一組治具，達成自動化彈性產線。

With the advancement of AI technology, intelligent robotic system have become a fashion trend. These systems are capable of processing human-like perception, allowing them to perform more complex tasks and objectives. In electronic factories, the automation of production lines for electronic components has become quite mature. However, for the assembly tasks of a variety of electronic connectors in small quantities, the cost of building automated production lines and vibration feeders is quite high. Therefore, this study aims to develop a grasping system using robotic arms, machine vision, and deep learning to adapt to the complex features and environment of micro electronic connectors. This system is integrated into the production line framework developed in this study, successfully achieving a flexible production line and address the high costs of traditional automated lines and vibration feeders, as well as labor shortages.
This paper uses the YOLOv8 object detection model and applies data argumentation techniques to train the model under varying lighting conditions with brightness adjustments, thereby enhancing robustness. The object detection model automatically generates multiple candidate bounding boxes. These are filtered by the Non-Maximum Suppression(NMS) method to remove redundant boxes. To improve the assembly success rate, this study designs an adapter fixture to arrange the pose of micro connectors. The gripper orientation is determined by calculating the direction of the opening based on the center of the bounding box and the image centroid. Since the assembly of the connectors requires a specific orientation, the object is placed with the opening facing up, down, or either the front or back, setting four distinct features for its pose within the adapter fixture. Then, the system waits for the signal from the rotary table to transfer the electronic connector from the adapter fixture to the insertion fixture. To enhance assembly accuracy, a calibration algorithm and calibration fixture are developed and integrated into the insertion fixture to determine the actual position of the assembly groove. Once completed, an IO signal is sent to the rotary table to confirm end the task, prompting the switch to the next fixture, thus achieving an automated, flexible production line.

[1] D. G. Lowe, "Object recognition from local scale-invariant features," in Proceedings of the Seventh IEEE International Conference on Computer Vision, 20-27 Sept. 1999 1999, vol. 2, pp. 1150-1157 vol.2, doi: 10.1109/ICCV.1999.790410.
[2] E. Rosten and T. Drummond, "Fusing points and lines for high performance tracking," in Tenth IEEE International Conference on Computer Vision (ICCV'05) Volume 1, 17-21 Oct. 2005 2005, vol. 2, pp. 1508-1515 Vol. 2, doi: 10.1109/ICCV.2005.104.
[3] H. Bay, A. Ess, T. Tuytelaars, and L. Van Gool, "Speeded-Up Robust Features (SURF)," Computer Vision and Image Understanding, vol. 110, no. 3, pp. 346-359, 2008/06/01/ 2008, doi: https://doi.org/10.1016/j.cviu.2007.09.014.
[4] E. Rublee, V. Rabaud, K. Konolige, and G. Bradski, "ORB: An efficient alternative to SIFT or SURF," in 2011 International Conference on Computer Vision, 6-13 Nov. 2011 2011, pp. 2564-2571, doi: 10.1109/ICCV.2011.6126544.
[5] A. Krizhevsky, I. Sutskever, and G. Hinton, "ImageNet Classification with Deep Convolutional Neural Networks," Neural Information Processing Systems, vol. 25, 01/01 2012, doi: 10.1145/3065386.
[6] R. Girshick, J. Donahue, T. Darrell, and J. Malik, "Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation," in 2014 IEEE Conference on Computer Vision and Pattern Recognition, 23-28 June 2014 2014, pp. 580-587, doi: 10.1109/CVPR.2014.81.
[7] J. Uijlings, K. Sande, T. Gevers, and A. W. M. Smeulders, "Selective Search for Object Recognition," International Journal of Computer Vision, vol. 104, pp. 154-171, 09/01 2013, doi: 10.1007/s11263-013-0620-5.
[8] R. Girshick, "Fast R-CNN," in 2015 IEEE International Conference on Computer Vision (ICCV), 7-13 Dec. 2015 2015, pp. 1440-1448, doi: 10.1109/ICCV.2015.169.
[9] S. Ren, K. He, R. Girshick, and J. Sun, "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 2017, doi: 10.1109/TPAMI.2016.2577031.
[10] J. Long, E. Shelhamer, and T. Darrell, "Fully convolutional networks for semantic segmentation," in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 7-12 June 2015 2015, pp. 3431-3440, doi: 10.1109/CVPR.2015.7298965.
[11] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 27-30 June 2016 2016, pp. 779-788, doi: 10.1109/CVPR.2016.91.
[12] C. Guo, X.-l. Lv, Y. Zhang, and M.-l. Zhang, "Improved YOLOv4-tiny network for real-time electronic component detection," Scientific Reports, vol. 11, 11/23 2021, doi: 10.1038/s41598-021-02225-y.
[13] G. Oh and S. Lim, "One-Stage Brake Light Status Detection Based on YOLOv8," Sensors, vol. 23, p. 7436, 08/25 2023, doi: 10.3390/s23177436.
[14] S. Yang, W.-T. Xiao, M. Zhang, S. Guo, J. Zhao, and S. Furao, "Image Data Augmentation for Deep Learning: A Survey," ArXiv, vol. abs/2204.08610, 2022.
[15] I. M. Chen and J. W. Burdick, "Finding antipodal point grasps on irregularly shaped objects," IEEE Transactions on Robotics and Automation, vol. 9, no. 4, pp. 507-512, 1993, doi: 10.1109/70.246063.
[16] J. Yun, S. Moseson, and A. Saxena, "Efficient grasping from RGBD images: Learning using a new rectangle representation," in 2011 IEEE International Conference on Robotics and Automation, 9-13 May 2011 2011, pp. 3304-3311, doi: 10.1109/ICRA.2011.5980145.
[17] I. Lenz, H. Lee, and A. Saxena, "Deep learning for detecting robotic grasps," The International Journal of Robotics Research, vol. 34, pp. 705 - 724, 2013.
[18] Y.-J. Chiu, Y.-Y. Yuan, and S.-R. Jian, "Design of and research on the robot arm recovery grasping system based on machine vision," Journal of King Saud University - Computer and Information Sciences, vol. 36, no. 4, p. 102014, 2024/04/01/ 2024, doi: https://doi.org/10.1016/j.jksuci.2024.102014.
[19] J. Terven, D.-M. Córdova-Esparza, and J.-A. Romero-González, "A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO-NAS," Machine Learning and Knowledge Extraction, vol. 5, no. 4, pp. 1680-1716, 2023. [Online]. Available: https://www.mdpi.com/2504-4990/5/4/83.
[20] S. Liu, L. Qi, H. Qin, J. Shi, and J. Jia, "Path Aggregation Network for Instance Segmentation," in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 18-23 June 2018 2018, pp. 8759-8768, doi: 10.1109/CVPR.2018.00913.
[21] T. Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, "Feature Pyramid Networks for Object Detection," in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 21-26 July 2017, pp. 936-944, doi: 10.1109/CVPR.2017.106.
[22] X. Wang and J. Song, "ICIoU: Improved Loss Based on Complete Intersection Over Union for Bounding Box Regression," IEEE Access, vol. 9, pp. 105686-105695, 2021, doi: 10.1109/ACCESS.2021.3100414.
[23] X. Li et al., "Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection," ArXiv, vol. abs/2006.04388, 2020.
[24] H. V. Koay, J. H. Chuah, C. O. Chow, Y.-L. Chang, and K. Yong, "YOLO-RTUAV: Towards Real-Time Vehicle Detection through Aerial Images with Low-Cost Edge Devices," Remote Sensing, vol. 13, 10/20 2021, doi: 10.3390/rs13214196.
[25] 陳尚宏，「整合光源控制與深度學習辨識之平放散料夾取系統開發」。碩士論文，機械工程學系，國立中央大學，2023。
[26] TM5-900產品型錄
https://www.tm-robot.com/zh-hant/tm5-900/，2023年7月
[27] TM-github
https://github.com/TechmanRobotInc/tmr_ros1/tree/noetic，2023年7月
[28] Robotiq
https://robotiq.com/，2023年7月
[29] ROS
http://wiki.ros.org/ROS/Tutorials，2023年6月
[30] Moveit
https://moveit.github.io/moveit_tutorials/，2023年6月
[31] Roboflow
https://roboflow.com/，2023年10月
[32] Camera Calibration and 3D Reconstruction, OpenCV
https://docs.opencv.org/4.x/d9/d0c/group__calib3d.html，2017年9月
[33] Ultralytics
https://github.com/ultralytics/ultralytics，2023年12月
[34] [Object detection] YOLOv8詳解
https://henry870603.medium.com/object-detection-yolov8%E8%A9%B3%E8%A7%A3-fdf8874e5e99，2024年1月
[35] R. Hartley and A. Zisserman, Multiple View Geometry in Computer Vision, 2 ed. Cambridge: Cambridge University Press, 2004.
[36] W. K. Pratt and J. E. Adams, "Digital Image Processing, 4th Edition," J. Electronic Imaging, vol. 16, p. 029901, 2007.