基於視覺伺服控制之機器人輸送帶追蹤技術及3D多異種物件抓取技術

Abstract

目前工廠自動化發展的瓶頸是執行任務間，人與機器的互動模式。機器人為了可以在生產線上快速地進行辨識和抓取，像人一樣具備高準確率之外，也要能夠感知外在環境發生的變化。在現行的機器人研究領域，物件追蹤技術在生產線裝配上已經屬於機械手臂的基本技術。傳統而言，機器人執行任務都是步驟進行，一旦第一個步驟失敗，將會導致接續的子步驟，難以進行。 現況而言，要在工業上解決這個問題，多半是依靠視覺系統的輔助為主。這是其中一種機器人的感知系統，機器人可以藉著他們的輔助做好事前準備，能更完美的完成任務。然後，每當無預期的事件發生時，傳統固定形式的步驟，由於不具備動態更新動作，不僅造成任務突然終止，也容易對環境造成不良的影響。即使依靠了視覺輔助，也僅能解決部分的突發狀況。 要解決這個問題的方法，一樣還是要依靠機器人視覺。在傳統情境中，輸送帶上的物體是靜止不動，機器人會根據命令進行物件抓取。當目標物開始在生產線上移動，任務的複雜性也隨之提升。這時使用視覺回授系統會是絕佳的解決方法。 因此本研究主題，提供了一個視覺辨識抓取兼追蹤系統，每次抓取姿態都是由回授系統來決定。這個系統架構分兩塊，各自擁有核心的演算法來進行物件追蹤。因為受限於環境的條件，整體實驗操作情形會有細節描述。此外對於追蹤時的抓取姿態，也有改良演算法去記錄補償值，有利於下一次的追蹤。

One of the bottlenecks for manufacturing automation is the interaction between robot and humans among tasks. Robots are not able to recognize the element from the assembly line quickly and accurately just like human operators do. Besides, in recent robotic field, conveyor tracking is one of fundamental function in the robot manipulator. Once this very first step fails in the production line, the latter subsequent operations are hardly to complete. Currently, the manufacturers solve this problem by using the vision in the environment. The robot arm now can perform the task and manipulate the work based on the precondition. Afterwards, something that may occur will be regarded as a kind of unpredictable circumstance among the task. Due to traditional static process, the condition may not only cancel the work but also have bad impacts on the environment. The problem is also about vision. In order to solve the dilemma, the practicable way is also based on robot vision system. In traditional scenario, the object keep stable pose on the assembly line. The manipulator follows commands to grasp the object. While the target is moving on the production line, the task becomes sophisticated problems. Thus, a distinct grasping method under visual control system is definitely one of the essential solutions. In this thesis, we propose a tracking strategy on moving objects for a robot arm object fetching system combined with distinct recognition algorithm. In addition, the grasping pose of robot arm is corrected by visual feedback system. The system is separated into two parts: eye to hand and eye in hand, and discussed in detail. Each part owns its core algorithm to complete industrial tracking and fetching tasks. Because of limitation from the environment, the working conditions will also be illustrated. Grasping pose for each type of element is adjusted by tracking and modification algorithms.