Path planning and sensor knowledge store for unmanned ground vehicles in urban area evaluated by multiple LADARs

The autonomous driving in urban areas has more problems to be solved than the off-road driving has. The research in this dissertation proposes the new methods for detecting road boundaries, saving information of the environment, and generating path that the vehicle has to follow for unmanned ground vehicles running in urban areas.;The algorithm to detect road boundaries is implemented by several methods to assign the confidential values to the results derived from the methods. The positions of the detected boundaries are transformed to the traditional grid map with the confidential values. The traversability values of the cells corresponding to the positions are adjusted by the confidential values.;The Sensor Knowledge Store to save the information of the surrounding of the vehicle uses the Quadtree data structure. The storage is very efficient for the usage memory and easy to search and update the nodes. The information that is saved in the storage helps the vehicle to generate an optimal path.;The new path planning algorithm is based on vector method that is different method from presented method. The path planner for the special behaviors, such as parking and n-point turn, are using Rapidly-exploring Random Tree (RRT) algorithm to find paths to perform the behaviors.;The research presented covers the theory of these new methods on an unmanned ground vehicle platform at the University of Florida's Center for Intelligent Machines and Robotics (CIMAR). Results have shown that the approaches in this dissertation are very useful for the autonomous driving in urban areas.