Distributed Sensing Testbed Development for Wavelet Based Global Map Estimation
The development of a fleet of flexible and ruggedized unmanned ground vehicles for use in autonomy and distributed sensing research has resulted in a mature platform with proven capabilities. Each Mapping Autonomous Ground Vehicle (MAGV) is capable of travel on- and off-road, speeds up to 10 mph, and its sturdy construction with a rugged suspension cushions onboard instruments from vibrations. The large battery capacity can sustain at least eight hours of hard use, including powering all onboard electronics. The MAGV is fitted with a high accuracy GPS/INS system for centimeter-accuracy localization and a powerful but compact onboard computer. The integrated wireless communications allow high-bandwidth data communication between the MAGV fleet and a base station. The platform can additionally be fitted with a wide array of sensors, including LIDAR and stereovision cameras, and is designed with ample space to allow the mounting of any future data gathering devices. The platform has already taken a central role in the development of new algorithms for map creation with modern sensing technology, and was deployed to collect data for the demonstration of the map estimation algorithms outlined in this thesis.
A wavelet basis combined with a state estimator is demonstrated to be effective for approximating a global map of a given area with complex features. The recursive least squares state estimator is highly effective at rejecting transient features, such as pedestrians frequently passing through the field of view, while retaining the shape of the walls and terrain features. The ability to vary the map resolution allows the mapping station to trade detail for a faster map update processing time. In its current implementation, the global map estimator supports the acquisition and integration of data from multiple simultaneous mobile sources. Because each scan is registered using the position of the vehicle when it is recorded, there is no difference between receiving all data from a single agent, or multiple agents working cooperatively gathering data in the same area. The wavelet basis also offers several opportunities for reducing communications overhead through data compression. In particular, we have demonstrated that simple thresholding of the least significant wavelet coefficients results in a significant reduction in data size with no noticeable reduction in fidelity of the reconstructed map estimate.