利用聲場特徵及光流影像定位之全方向運動平台

Abstract

本論文提供在室內環境中的機器人一套完整且低成本的定位方法，其整合了聲場特徵定位法以及光流影像定位法。聲場特徵定位法是由個人電腦端的雙聲道麥克風收錄機器人在不同位置所發出的聲音資訊，並依據各位置上收錄聲音的相位差以及大小比分佈，預先建立高斯混合聲場地標模型；使機器人在模型環境中移動並發出聲音時，可藉由將新的聲音資訊代入模型中運算，以偵測機器人所在的絕對位置。而光流影像定位法是由兩個光流感測器對地面偵測各別的位移資訊，並加以演算合併以求得機器人的相對位移資訊。在此更進一步地利用機率格方法整合以上兩種定位資訊，以提升絕對位置定位的可靠度。 本論文亦提出以全方向運動為基礎的機器人平台，相較於傳統輪式平台，此平台具有較高的運動靈活度；並配合嵌入式網路技術加以整合，達成分散式運算及遠端操控機器人的目的。

This thesis proposes an integrated and low-cost method of localization, which can estimate the absolute location of a mobile robot in a complex indoor environment by using sound field characteristics and optical flow sensing. The first method, Localization using sound field characteristics, record the sound information by two channel microphone of personal computer, and establish Gaussian mixture-sound field landmark model according the phase difference and magnitude ratio distributions between two microphone at each landmark. When the robot moves and generates sound in the modeled environment, we can bring the new sound information into the landmark models to detect it’s location. The second method, Localization using optical flow sensing, use two optical flow sensors to estimate the relative movement information of the mobile robot. Moreover, we utilize the position probability grids method to integrate these two methods for improving the reliability of estimating the absolute location. In this thesis, we also propose the omni-directional platform offer a higher mobility and the embedded Ethernet technology is used to achieve the objectives of distributed computation and remote control.