應用於孔裝配任務包含物件整理的基於視覺與幾何關係之易校正的自動機器多手臂系統

Abstract

工業機器人對於加工和工業自動化是必不可少的，但目前的系統多不夠靈活，無法應對較短生命週期的產品。為了降低成本，裝載工件的夾具應該從生產線中移除。本研究旨在設計一個容易校準、無需預設軌跡的自動化系統，以提高靈活性以適應更多可能的產品。 在這篇論文中，一個配備了RGB-D和RGB攝像頭的多臂機器人裝配系統被提出，用於自主的工件和插件整理以及孔裝配任務。該系統能夠自動定位工件，整理它們，然後提供孔裝配的動作來組裝工件和插件。該系統在初次使用需要一系列的校準，以協調系統不同部分之間的軌跡規劃，並將相機傳感與其他視角相關聯。其次，通過提出的基於點對特徵（PPF）和迭代最近點（ICP）的姿態估計方法和提出的SpinCheck來估計工件的6D姿態，以便在高噪的點雲下進一步提高準確性。接著，主互動手臂將重新抓取它們，直到它們達到理想的姿勢。而後，因為插件太小以至於無法在點雲中被感測，我們將通過簡單的二維圖像估計來整理插件。接下來，通過二維圖像估計以提高精度，將這兩個工件對齊並放在一起。最後，利用二維圖像進行一系列的對齊控制，包括六邊形和矩形的幾何關係和一些策略，以克服基座校準誤差。 最後，我們通過實驗評估了所提出方法的性能，並驗證所提出的系統的實用性。對實驗數據進行了分析。

Industrial robots are essential for machining and industrial automation, but present systems are not flexible enough to handle shorter life cycles. The need for specific fixtures should be removed to reduce costs. Designing an automated system that can be easily calibrated without pre-defined trajectories will increase flexibility to adapt to more possible products. In this thesis, a multi-arm robotic assembly system equipped with an RGB-D and a RGB camera is proposed and delivered for autonomous workpiece and plug organization and peg-in-hole assembly tasks. It can automatically locate the workpieces, organize them, and then deliver the peg-in-hole motion to assemble the workpieces and plugs. Initially, the system needs a series of calibrations to set up coordinates between different parts of the system for trajectory planning and relating camera sensing to the other perspective. Secondly, the 6D pose of the workpiece is estimated by the proposed pose estimation method based on point pair features (PPF) and iterative closet point (ICP) and a proposed SpinCheck for accuracy under noisy point clouds. Then the main interactive arm will regrasp them until they attain the desired pose. Later, the plugs will be organized by simple 2D image estimation since plugs are too small to be sensed in a point cloud. And the two objects will be aligned and put together by 2D image estimation for higher accuracy. At the end, a series of alignment control with 2D image involving the geometry of hexagon and rectangle with some strategies to conquer errors in base frame calibrations. Finally, experiments are carried out to assess the performance of the offered approaches and confirm the practicality of the proposed system. An analysis of the experimental data is offered.