漸進式運動計畫之街圖管理

• Main Authors: 謝揚權, Hsieh, Yang Chuan
• Language: 中文
• Published: 國立政治大學
• Online Access: http://thesis.lib.nccu.edu.tw/cgi-bin/cdrfb3/gsweb.cgi?o=dstdcdr&i=sid=%22A2010000153%22.

　　運動計畫的技術已被廣泛地應用在機器人學及電腦圖學等領域。其基本問題的型態，是在一散佈著障礙物的場景中，給定物體之起點與終點，再利用運動計畫的演算法，來搜尋該物體由起點到終點不會碰撞到障礙物的可行路徑。這類問題的運動計畫方法可分為兩類：單一查詢及多次查詢。前者的好處是可以應用於動態的環境中，而後者的優點是可以透過對環境做事前的處理而減少搜尋所花費的時間。在本論文中，我們延展文獻中快速擴展隨機樹RRT (Rapidly-exploring Random Tree) 的結構，建立一種稱為RRF (Reconfigurable Random Forest) 的資料結構，用來有效解決運動計畫之問題。此方法是以漸進學習的方式，逐步地在場景中建構出運動計畫所需之街圖，並運用一些精簡街圖的策略，可以有效地管理維護街圖之成長。RRF同時具備了單一查詢及多次查詢的優點，我們分別於靜態以及動態的環境下實驗，以觀察其實際運作的情形，並針對精簡街圖之有效性與影響作深入的探討與評估。實驗結果顯示，此方法可有效提昇運動計畫器的效率及其適用的範圍。 === 　　The motion-planning techniques have been widely applied to many domains, such as robotics and computer graphics. The basic problem of motion planning is about finding a collision-free path for a robot, moving in a workspace cluttered with obstacles. Traditional approaches to the motion-planning problem can be classified into single-query and multiple-query problems with the tradeoffs on run-time computation cost and adaptability to environment changes. In this paper, we extend the Rapidly-exploring Random Tree (RRT) structure proposed in the literature, to a more flexible data structure, called Reconfigurable Random Forest (RRF). This approach can learn incrementally on every planning query and effectively manage the learned roadmap. This planner is as efficient as other single-query planners and the performance gets improved when the learning process goes on. Our experiments show that the planner can also account for environmental changes and maintain a concise and representative roadmap. Experimental results show that this planner has broadened the applicability of motion planners to problems of different characteristics and complexities.