使用分散式計算之室內環境探索機器人

Abstract

本論文提出一台以機器人作業系統(ROS)架構下 Gmapping結合環境探索和路徑規劃之機器人的設計與實現，使機器人能在未知環境中自主運用環境探索演算法、路徑規劃演算法和運動學控制器探索環境，並在探索環境的同時使用 Gmapping 建置二維平面地圖，使之完成自主探索環境並建置地圖的功能。環境探索演算法中使用的是本論文提出的「分群式邊界偵測法」，使機器人能先把周圍環境探索完畢後才繼續探索新的環境。而路徑規劃使用的的是本論文提出的「改良權重A*演算法」，能使機器人避開障礙物且規劃完整的路徑。由於本實驗室想設計一個輕量化的設備來取代笨重的電腦，但速度卻不會比電腦慢太多的方法，所以採用機器人作業系統(ROS)架構實作在多塊 Udoo Quad 板上，其分散式架構剛好符合要求，能把系統的架構分散開來，使單一程序可以完全使用一個 Udoo 板的效能。為了驗證系統的性能與效能，本論文利用室內環境進行諸多實驗，而由實驗結果可知，本論文設計的機器人導航系統確實能達到分散式處理之輕量化之導航與探索功能的目的。

In this paper, a system integration of Gmapping, environmental exploration and path planning is provided based on Robot Operating System (ROS) and multiple Udoo single board computers in order to develop a mobile navigation robot. As such, the robot is capable of navigating in an unknown indoor environment using light-weight, low-cost, and high-computing platforms. To estimate the robot pose as well as building a top-view map, Gmapping is used due to its robust performances. Moreover, a clustering and frontier-based exploration method is proposed such that the robot is able to efficiently explore nearby environments before moving to newly observed ones. As for path planning, an improved A* algorithm is proposed to provide an optimal path without colliding with obstacles. To effectively operate the overall navigation system, a distributed computing system using ROS is designed so that algorithms can perform in parallel without decreasing the efficiencies of Udoos. To verify the performances of the proposed navigation system, various experiments are conducted, and experimental results show that the robot can reliably navigate and explore an unknown indoor environment.