應用基於視覺與力覺之監督式遠端遙控於工業巡檢中的機器人靈巧操作任務

Abstract

於工業廠房中，需維護的設備種類繁多，傳統倚賴人工巡檢之方式不僅耗時，亦易受限且缺乏效率。因此，自主式機器人系統被視為可行的解決方案。本論文建立一套監督式遠端操作系統，以執行工業巡檢場景中的靈巧操作任務，例如按鈕與閥門之操作。為解決此類問題，本論文提出一種結合監督式控制 (supervisory control) 與局部自主性 (local autonomy) 的控制架構，兼具全遠端操作與全自主系統之優勢。使用者透過無線通訊控制遠端機器人執行任務，並可監督系統之自主運作流程，必要時亦可即時介入。此架構有助於減少人力監督負擔，同時提升任務之成功率。於局部自主性流程中，系統採用六自由度物件姿態估測與追蹤，以實現漸進式 (coarse-to-fine) 的視覺控制策略。在操作物件前，透過力導引自動對位（force-guided self-alignment）機制，進一步調整最終夾持姿態。實驗結果顯示，本論文所提出結合視覺與力覺感知模態之控制架構，能有效修正靈巧操作過程中的夾持姿態誤差，並於特定軸向達成小於 1.5 [mm] 及 2.6 [deg] 的誤差範圍。

In industrial plants, a wide variety of equipment is to be maintained, and hence manual surveillance tours are scheduled, which is not only time-consuming but also limited and effectless. Autonomous robot system is a viable solution to address this limitation. In this thesis, a supervisory teleoperation system is set up to perform dexterous manipulation tasks in industrial surveillance scenarios, such as buttons and valves operations. A supervisory control scheme with local autonomy is proposed to address this problem, which combines the advantages of full teleoperation and full autonomy. The user controls the remote robot to perform tasks through wireless communication link, and supervises the local autonomy and can intervene the tasks anytime if needed. This helps reduce human supervision efforts and at the same time increases task success rates. In the pipeline of local autonomy, the object 6D pose estimation and tracking are employed for a coarse-to-fine vision-based control framework. A force-guided self-alignment scheme is used to adjust the final grasp pose before operating the object. Experimental results show that the proposed control scheme with vision and force-torque (F-T) modality is able to adjust the grasp pose for the dexterous manipulation task with errors less than 1.5 [mm] and 2.6 [deg] in the selected axes.