以類神經網路應用於機器人室內定位之研究

Abstract

近年來有關機器人的研究逐漸受到重視，例如室內導覽、居家看護及環境探測等應用。其中有許多應用必須知道機器人的位置，因此有許多不同方法來偵測機器人所在的位置。全球定位系統(GPS)是最普遍的應用系統，但因為全球定位統在室內環境中會受到建築物屏障效應的影響，而無法有效的應用在室內環境中。本研究目的即在利用低成本與維護方便的 ZigBee 系統與感測網路實現機器人室內定位系統。 先前有些室內定位系統使用最大概似估計（Maximum likelihood estimation，MLE）演算法來作定位，但因為感測訊號易受干擾，造成定位上的誤差。因此我們利用類神經網路，以倒傳遞演算法(back-propagation network，BPN)實施機器人的定位。操作上容易取得基地台間的訊號強度(RSSI)訊號，且在定位誤差方面，有顯著的改善。經過實驗，發現在10×7m室內環境中，以4個參考節點定位最有效。雖一般認為節點越多定位越精準，但因訊號重疊與干擾嚴重，並不需要放置過多的感測器。 本研究的結果可供改進室內定位系統設計參考。Recently people paid more attention to the research of robot such as indoor guide, personal security caring and environment monitoring. Among them, the positioning technique is required for controlling robot moving. Many different approaches have been proposed to tackle the problem of determining the robot position. In an outdoor environment, the Global Positioning System （ GPS ） is the most popular approach. However, due to the poor indoor coverage, the GPS cannot provide a satisfactory solution to the problem of indoor location estimation. The purpose of this research is to develop a low cost and practical system with Zigbee system to implement indoor robot positioning with sensor network. Some other indoor positioning systems used Maximum likelihood estimation (MLE) algorithm to deal with reception signal. But the positioning accuracy was easily disturbed by interference noise. We then used back-propagation neural network (BPN) to improve the accuracy of positioning because its signal strength indication （RSSI）value was easily to receive and the error could be reduced. After several times experiment, the BPN-based algorithm did much better performance than MLE-based indoor positioning. It was found that the best condition was 4-reference node in 10x7m indoor environment. We cannot say as usual that more sensors can do much better for positioning because it will cause much signal overlapping and interference. The results of this research can be used as reference for indoor positioning design.