國立臺灣大學應用力學研究所王立昇2006-07-262018-06-292006-07-262018-06-292000-11-30http://ntur.lib.ntu.edu.tw//handle/246246/21589To perform the navigation of an unmanned vehicle, various sensors, such as GPS, INS, compass, encoder, etc, can be used. Depending on their characteristics, different sensors may have different advantages. For example, GPS may be more sensitive to low-freqency noise, while INS is more susceptible to high-frequency noises. In order to integrate these sensors, the algorithm of data fusion along with the Kalman filter may be adopted. However, there are some issues having to be tackled. First, the initial setting of the algorithm must be given. Secondly, if the assumption of independence in the Kalman filtering is not valid, it is necessary to deal with dependent processes. In this report, an algorithm of determining the initial settings, including the error covariance, the process noise covariance and the measurement noise covariance, is proposed. On the other hand, the covariance intersection algorithm is used to solve the problem regarding the dependence of the information. The combination of these strategies is then used to design the fusion INS-GPS system for the navigation of a vehicle. The experimental results showed that the algorithm is more robust comparing with classical Kalman filtering algorithm.application/pdf236841 bytesapplication/pdfzh-TW國立臺灣大學應用力學研究所reporthttp://ntur.lib.ntu.edu.tw/bitstream/246246/21589/1/892212E002084.pdf