機器人整合3D物體辨識與夾取系統應用於工廠自動化

Robot Integrated 3D Object Recognition and Fetching System for Factory Automation

郭佳文(Chia-Wen Kuo)

指導教授 ： 羅仁權

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.00150

摘要

目前工廠自動化發展一個重大的瓶頸，在於機器視覺，機器人沒辦法像產線上的工人一樣，能夠快速且精確的辨識出從輸送帶運輸過來，任意擺放的複雜物件，既然無法辨認出物件，那就更遑論如何將物體抓取起來，進行後續的操作像是組裝、焊接、塗膠等等操作。
為了解決這個問題，目前大多數的廠商是想辦法將物件的種類和位置固定，捨去辨識場景中物體的步驟，轉而依賴機器手臂高精準度的特性，在已知位置之間來回運動和操作。這樣一來一但物件位置有偏差，就會導致災難性的後果，而且也白白花費了高額成本在精準地擺放物體。
因此本研究的主題，在於如何將三維物體辨識整合到這個系統中，讓機器手臂可以自動辨識場景中要操作的物體，並且結合直覺性教導的功能，來事先教導手臂如何正確且穩定的抓取辨識出來的物件，當中最重要的兩個模組，分別是物體辨識，以及機器手臂本身的控制。
在本研究中，我們成功的實作了一個整合式的系統，來達成物體自動辨識與抓取，我們也對物體辨識和抓取進行大量的測試，並識別出整個系統的瓶頸，以利後續研究者能夠基於此系統繼續發展。

關鍵字

工廠自動化 ； 生產自動化 ； 機器手臂 ； 三維物體辨識 ； 物件抓取

並列摘要

One of the bottlenecks for manufacturing automation is machine vision. Robots are not able to recognize randomly oriented components coming from the assembly line quickly and accurately just like human operators do. Once this very first step fails, any other subsequent operations such as picking up the component, assembling, welding, painting, etc, are impossible.
Currently, manufacturers solve this problem by fixing the component. The robot arm then performs the task and manipulates the component based on this precondition. This approach totally omits the fragile object recognition step and relies solely on the precision and repeatability of the robot arm. Once there are pose error setting up the component, a disastrous consequence may occur and the whole manufacturing process might shutdown simply because of this minor fault.
The research objective is to integrate 3D model-based object recognition into the system for the capability of the robot arm to recognize the component in the scene. Furthermore, teaching by touching is integrated to let human operators teach the robot how to pick up the components stably. Two of the most important modules for the success of this integrated system are 3D object recognition system and the manipulator itself.
In this research, we successfully implement an integrated system for recognizing and fetching the randomly oriented objects. We also evaluate the system extensively and identify the bottleneck of this system, hoping that this could open up a road for robot-integrated manufacturing automation and become the basis for future research.

並列關鍵字

factory automation ； manufacturing automation ； 3D object recognition ； object fetching

參考文獻

[1] Kinova JACO2 is available at http://www.kinovarobotics.com

[4] C. Loughlin, A. Albu-Schäffer, S. Haddadin, C. Ott, A. Stemmer, T. Wimböck, et al., "The DLR lightweight robot: design and control concepts for robots in human environments," Industrial Robot: an international journal, vol. 34, pp. 376-385, 2007.

[11] K. B. Shimoga, "Robot grasp synthesis algorithms: A survey," The International Journal of Robotics Research, vol. 15, pp. 230-266, 1996.

[12] Z. Zhang, "Microsoft kinect sensor and its effect," MultiMedia, IEEE, vol. 19, pp. 4-10, 2012.

[14] Reflexxes Online Trajectory Generation is available at http://reflexxes.ws