跨樓層文件傳遞機器人之設計與實現

Abstract

本論文提出一擁有跨樓層文件傳遞功能之機器人的設計與實現，使輪型機器人能夠在一已知地圖的大樓中自主導航，並搭配影像及機械手臂，使機器人能夠搭乘電梯上下至目的地樓層，再配合開發於手機上的應用程式，讓使用者可以藉由使用者介面傳遞目的地資訊給機器人，使其前往目的地。針對定位演算法，本文提出「嵌設錯誤修正向量之蒙地卡羅定位法」，使傳統的蒙地卡羅定位的感測器資訊不僅僅只用來判別粒子的好壞與否，更可進一步得知下一刻時間的參考向量。路徑規劃部分本文提出了「改良型A*混合多重骨架路徑規劃演算法」，以改善傳統骨架繞路的問題，並搭配搜尋上下載點的策略，使其規劃出一條遠離障礙物並安全的路徑，不管在路徑長度還是規劃時間都較傳統A*混合骨架演算法來得優異。在電梯按鈕辨識部分，本篇論文使用輪廓提取的方式，對建立好的模組進行比對，使機器人得知電梯按鈕的座標。手臂控制方面，主要是搭配單攝影機，將三維正逆向運動學的數學模型簡化為二維，可使較不精準的機器手臂如同人類的手臂一樣，朝向按鈕伸直，並觸碰按鈕。最後再將使用者介面實現於Android智慧裝置上，搭配TCP/IP通訊，以及語音辨識工具，讓使用者可以用簡單的使用手機應用程式，命令機器人前往目的地。本論文最後以多個實驗結果驗證所提出之方法的可行性。

This paper proposes a design and implementation of a document delivery mobile robot, which is capable of navigating in a known environment and taking elevator by a webcam and robotic arm. By developing a voice recognition program, users can order the robot to move to a destination by using a mobile device. Regarding robot localization, this thesis proposes a Monte Carlolocalization (MCL) algorithm incorporating an error correction vector, which uses the sensor measurements not only to obtain weights of particles, but also calculate an error correction vector to improve the overall localization accuracy. Moreover, a hybrid path planning algorithm using an enhanced multi-skeleton and A* approach is employed in this thesis to obtain a favorable path for the robot, which extracts a multi-skeleton of the map and introduces a strategy of searching preferable uploading points. To take an elevator reliably, a connected-component approach is used to match the model of each button on the elevator panel so that the robot can detect the buttons successfully. After detecting the buttons, kinematics is employed to control the robotic arm to push the button. Finally, the performances and the feasibility of the proposed document delivery robot are confirmed by several experimental results. Keywords: Monte Carlo Localization, A Star, skeleton, navigation, mobile robot.