| There has been an increase in the usage of sensor technology as they are adaptable for use in different environments, many of which are hostile to direct human observation, such as regions affected by land mines or forest fires. The sensors used to monitor developing situations are small inexpensive computing devices with limited processing capabilities, limited power supply and may be destroyed in the event that they are monitoring, without suffering a great financial cost. Also, since these devices are inexpensive, multiple sensors can be deployed for the same application where the sensors are placed or attached to a mobile platform in a particular configuration or shape. However, because these devices are inexpensive, they do not possess all the software and hardware necessary to produce accurate observed data from the environment they are deployed in. Moreover, there are additional factors such as atmospheric conditions, network delays, sensor characteristics etc. that may affect the measures being monitored and therefore, the data observed and produced as output by the sensors may be inaccurate.;The issue of obtaining accurate estimates of location and orientation from inaccurate observed sensor data has been subjected to thorough investigation in the literature. However, in the application environment of multiple Global Positioning Satellite (GPS) sensors attached to a mobile robot platform, these previous methods do not take advantage of the sensor configuration information to produce more accurate estimates of the measures being observed. In this dissertation the authors will demonstrate that in fact, with the use of the sensor configuration and the inaccurate observed sensor data, it is possible to obtain accurate estimates of location and orientation of the deployed sensors.;In this dissertation, we propose several concrete issues and their respective solutions for the framework of the mobile robot platform with GPS sensors attached in a particular configuration to be operational. We use optimization techniques to fit estimates of locations and orientations of the sensors on the mobile platform given observed data that is highly unstable when the platform is stationary or in motion, while taking advantage of the known configuration knowledge. To deal with outliers and missing data, we use statistical and heuristic weighting techniques to favor accurate observed sensor data over inaccurate data. Moreover, the production of estimates through the use of optimization has to conform to the real-time constraints when the platform is in motion and therefore, we introduce sliding windows to be able to generate updated estimates. Furthermore, depending on the size of the sliding window, the task of correcting the lag between the estimate over the sliding window and the estimate with respect to the current time-step is also considered to produce more accurate results. |
