室內構造化環境中藉由圖形比對演算法與感測資訊融合達成廣域地圖建立與行動機器人姿態估測

Abstract

近年來，機器人已被廣泛的使用在各種用途上，例如軍事行動、人機互動、保全系統或者是環境探勘等各方面的任務，都是非常典型的應用。而裝備有測距儀的輪型行動機器人是經常被使用的平台之一。 本篇論文採用裝備有雷射測距儀的雙輪式的行動機器人做為理論基礎以建立模型，試圖達成在室內結構化之環境下機器人建構地圖與姿態估測的應用，機器人在室內環境中進行導航時，所使用之感測器可被粗略分類為外部與內部感測器，而內部感測器所感測之資訊會因雜訊或其他因素導致與實際狀態產生誤差，例如輪胎的打滑狀態會使編碼器對機器人的運動軌跡做出錯誤的估測。 為了處理上述之現象，在此時需要引入外部感測器的資料，雖然外部感測器也會有雜訊的干擾，但適度地結合內外部感測器可對內部感測器感測不精準之現象做出修正。 在本論文中，機器人主要的外部感測器為雷射測距儀，所得資訊主要為周遭障礙物之距離資訊，在機器人移動時，這些資訊會有一定程度的重複性，而利用最小方差法概念的迭代最近點演算法，在圖形資料具有相當程度的相似性時，於圖形比對及疊合上能有相當程度的效果，所以對於地圖的建立與姿態估測來說，迭代最近點演算法經常能發揮相當程度的貢獻。 然而使用圖形比對演算法無法避免的問題是誤差的累積，本論文從各種角度預先分析環境資料之圖形點，利用各種演算法進行資料的過濾，以降低迭代最近點演算法之計算誤差與時間消耗，再利用感測器融合之概念進行資訊整合，此方法可同時對機器人姿態及整體地圖做出校正，最終使機器人經運算得到完整的室內地圖及取樣過程中相對精確之姿態資訊。

In recent years, mobile robot has been widely used in many applications. The wheel mobile robot (WMR) that equipped range finders is a platform which is commonly used for some purposes such as military operations, human-robot interactions, security systems and environment explorations. In this thesis, a model of two-wheel mobile robot with laser range finder is constructed doing pose estimation and map building about the robot in a constructed indoor environment. The sensor could be roughly classified as internal sensor and external sensor in robot navigating problem. But the internal sensor may have some sensing error or noise. For instance, the skidding and slipping phenomenon between the wheel and ground would lead to a wrong estimation for the robot trajectory. In order to handle these errors, the external sensing data should be taken into consideration. Although the external sensor also has the noise effect, the combination of both sensors could make a modification of the mentioned results above. The main external sensor is laser range finder in the thesis, and the obtained information is the distance data of the obstacle around the robot. These data would have the similarity in constructed environment. The Iterative Closest Point algorithm, called ICP algorithm, using the concept of least square error and closest point could be used to solve problems like data comparison and shape matching. It could provides a considerable improvement about map matching and obtain translational and rotational displacements between two sampling point of the mobile robot. However, error accumulation is an inevitable result in simulation and experiment. The issue of doing preprocess to collect environment data is discussed in this thesis. These data could be filtered by several algorithms so the error could be eliminated and lower down the time consuming in the computing procedure. Finally, the data fusion by using the concept of sensor fusion could give a correction of robot pose and build a global map.