可適應無人搬運車彈性化設計之學習式導航策略及強健式路徑跟隨控制

Abstract

現今無人搬運車(Automated Guided Vehicle，AGV)引入製造工廠和自動化倉儲是邁向工業4.0的必備條件，由於實際工廠生產線環境中高度動態與不確定性，本論文開發一套強化AGV定位精確性與導航策略。首先提出具有低成本效益之反光柱輔助定位技術，利用反光點作為環境中的分離特徵進行重新定位，能有效改善自適應蒙地卡羅定位定位(Adaptive Monte Carlo Localization, AMCL) 演算法在環境特徵不明顯或環境地圖邊界過於破碎，所導致的迷航或定位失效的問題。接著，本論文提出可適應AGV動作的路徑跟隨控制設計，並整合至機器人作業系統(Robot Operating System, ROS)的軟體環境，此種設計除了可延伸應用於相關自主式無人搬運車軌跡追蹤控制策略之外，基於模糊神經網路架構並提出新的誤差計算方式，可以在模擬環境搭配AGV運動模型來預先進行控制參數自動調整。本論文開發的AGV導航控制先使用MATLAB模擬環境來實現所提出的用於導航控制的模糊神經網絡(Fuzzy Neural Network, FNN)策略，對軌跡跟踪中的模擬結果評估，以驗證所提出的AGV控制策略的有效性。由實驗測試結果說明，本論文提出的反光柱輔助定位搭配AMCL定位演算法能有效克服累積定位誤差之外，進一步整合強健式路徑跟隨控制與學習式導航策略，能展現本論文所開發AGV技術在實際工廠生產線環境中之高度應用價值。Nowadays, automated guided vehicles (AGVs) have become a bridge for manufacturing factories and warehousing industries to enter Industry 4.0. This study focuses on the development of sophisticated localization algorithms and navigation algorithms based on AGVs. First, a reflector-assisted localization method is proposed, which is different from the Adaptive Monte Carlo Localization (AMCL) algorithm used by most mobile robots. With the installed reflectors, this method improves the problem of AMCL localization when the environment features are not obvious or the boundaries of map are broken, which may lead to the problem of getting lost its position or localization failure. The current position of the AGV is calculated and localized using reflection points as the second feature of the environment, especially while the AMCL localization fails and the supported localization information from the reflectors can be supplemented in time. In addition, a path following algorithm using intelligent control design is proposed and integrated into the robot operating system (ROS). Based on the fuzzy neural network (FNN) approach, such a proposed intelligent controller can be used in the pre-built virtual environments and the required parameters of FNN can then be determined. The developed control system uses a MATLAB simulation environment to implement the proposed FNN approach in the navigation task of AGVs. The simulation results with the real experiments are evaluated to verify the effectiveness of the proposed navigation and control strategy. To sum up, the demonstration results depict our approach, including the reflector-assisted localization method and path-following control strategy, with the high potential for the production lines of factories.