自走機器人應用樓層平面圖資訊於未建構環境下實現導航及定位

Abstract

本論文所討論的是一種新的建立地圖及機器人導航的演算法，我們稱之為直覺式地圖擷取導航演算法(Intuitive Map Extraction Navigation algorithm, IMEN algorithm)。使用的機器人搭載有三種感測器：超音波陣列、視覺攝影機、及馬達的編碼器。機器人可以利用它們自主地辨識一個未探索過環境的樓層平面圖，並往目的地前進。首先，機器人透過攝影機擷取環境中已知位置的平面圖影像，透過影像處理、樣板比對演算法(Template Matching)、支持向量機(Support Vector Machine)來取得未探索環境的格子地圖(Grid Map)及各房間的相對關係，再透過修改過的A*演算法做路徑規劃，並規劃往操作者給定的目的地前進的路徑。其中，如何將與現實環境中不成比例的樓層平面圖化為機器人可用的機器人地圖，並藉由不準確的相對特徵點間關係來規劃路徑並進行導航及定位機器人的技術，是本論文探討的最大挑戰處。在實驗結果中，我們完成了門牌辨識及地圖辨識的演算法，並成功地導航機器人至目的地處，與傳統利用雷射測距儀花費長時間建立地圖的演算法相比，利用有領先知識的樓層平面圖，此演算法可以在一~兩分鐘之內完成地圖建立及路徑規劃。

The thesis discusses a new algorithm about robot mapping and navigation. We call it IMEN (Intuitive Map Extraction Navigation) algorithm. The experiments utilize robot platform equipped with 3 kinds of sensor, camera, ultrasonic array and encoder. Robot can recognize the floor plan map of unstructured environment and navigate itself to the destination. Under the process, robot can also localize the pose of itself. First, it extracts floor plan image from camera for getting grid map and features by image processing and template matching algorithm. Then, modified A* algorithm is used for path planning. The most challenge problem is that how to extract map and plan path from unproportionate floor plan map and utilize them for localizing robot. In our experimental results, we finished the algorithm of room plate recognition and map recognition. Then, robot can be navigated to the destination successfully. Compared with the traditional mapping method of SLAM which takes long time, robot uses prior knowledge of floor plan map of unstructured environment. Algorithm can build grid map and path planning 1 ~ 2 minuts.