基於權重式連續型ICP演算法之雲端建圖

Abstract

本篇論文係透過機器人搭載雷射測距儀，連續地搜集環境中機器人與障礙物之間的距離資訊，並將距離資訊轉換為座標集合，再藉由迭代最近點演算法(Iterative Closest Point, ICP)將座標點集合疊合為一張地圖。由於傳統迭代最近點演算法容易受到資料的初始位置影響而造成建圖錯誤，且隨著建圖進行，資料量逐漸增加也會導致了運算時間的增長。故本論文提出ㄧ權重式連續型ICP演算法，利用機器人在連續搜集距離資訊時，相鄰兩筆資料差異不大的特性，使用相鄰的資料進行ICP演算法，再利用前幾組座標集合計算出的旋轉量、位移量以修正當前集合的位置並與全域地圖重合，藉由更新當前集合再修正集合位置的方法，避免了因初始位置差距過大而導致的建圖錯誤，同時降低了與全域地圖運算的資料量，也透過兩次的ICP運算，提出了改良式權重連續型ICP演算法以進一步提高運算的精準度。最後將此序列式演算法，透過雲端架構修改為分散式平行架構，將ICP演算法中大量使用的迴圈運算的步驟分散給雲端叢集處理，以降低運算的負擔，提高運算效能。

This paper proposes a method of map building based on cloud-computing with a weighted iterative closest point (ICP) algorithm, in which a laser range finder is used to measure the distance between the robot and obstacles. Previously, after converting these data into cloud point sets, the traditional ICP could be used to align two of these sets. However, two problems exist with this method. Firstly, its efficiency and accuracy can be affected by outliers and noises. Secondly, the model set continues to grow during the ICP process. Therefore, this paper proposes continuous alignment to solve these problems. We take one cloud point set and another set of the next generation to build the map; because these sets share a higher similarity than the sets used in the traditional ICP, higher accuracy can be achieved. Besides, as the continuous ICP proposed by this paper has a fixed reference set, in contrast to the traditional ICP with a growing set, time consumption is also improved. In addition, a modified continuous ICP algorithm is introduced to reduce errors. Finally, cloud computing architecture is used to realize parallel computing for reduction of computational burdens.