基於Kinect感測器之智慧型履帶機械人 於未知坡道環境行走設計

Abstract

為了使履帶機械人在未知環境中可以有效克服特殊的地形，本論文提出利用Kinect感測器為基礎之履帶機械人的控制方法，其中控制模式包括尋找模式、對準模式、近距離模式及爬坡模式。目前大部份智慧型機器人在未知環境中尋找目標物或辨識物體需靠CCD去辨識，然而在顏色複雜環境下會導致辨識效果大幅度降底，本論文利用Kinect感測器抓取出目標物的深度資訊來進行物體辨識，利用深度影像灰階圖去辨識目標物，可改善傳統CCD在影像辨識上的光線影響，並且可引進新穎的深度辨識概念。傳統機械人在不同控制模式間的轉換都以階段式為主，階段式的控制方法若在某個模式中發生問題，則在後面的模式將無法執行，所以在本論文中引進了第二類區間式模糊系統設計(interval type-2 fuzzy fusion)來平行整合不同控制模式，第二類區間式模糊系統會即時進行調整每項控制模式的權重值，調整權重後，各項控制模式會按照權重要性決策輸出。最後在本論文中，我們利用在未知環境中搜尋斜坡並完成爬坡的實驗來驗證本論文的行為模式和智慧型機器人的效能。To achieve the goal of overcoming the special terrain in an unknown environment, several control modes based on a single Kinect sensor for a tracked robot are proposed in this thesis. The control modes include a search mode, an alignment mode, a closing mode and a climbing mode. In general, most of intelligent robots must be relied on CCD sensors to find the target or identify objects. However, measuring color changes frequently when robots place in the complex environment and it leads to the difficulty of identifying the objects. In this thesis, according to depth information of the measured objects provided by Kinect sensor, the track robot can recognize objects without considering the change of light. Two kinds of decision method are used to change the modes when the track robot is during performing specific tasks. One is the traditional staged decision method, which is easy to program but is low reliability, and the other is the interval type-2 fuzzy decision method, which can smoothly change mode because of the ability of parallel computation and system uncertainties involved. Finally, the slope climbing experiments in an unknown environment are used to show performance of our proposed control method for the track robot.