In 2007, the CMU team won The DARPA Urban Challenge with the autonomous vehicle entry "Boss". The urban navigation components (e.g. motion planner) depended heavily on pre-stored map information and world modeling. A Road Network Definition File (RNDF) was used to store the road information.
The most basic component of a RNDF file is the road lane information, which is encoded by a sequence of lane waypoints. Rather than empirical (manual) specification, the choice of the waypoints should be related to both road geometry and the interpolation method (used to reconstruct a finer road representation to have smooth interpreted curvature). Furthermore, the road lanes may branch out or merge together occasionally. A "virtual" lane is used to join an exit lane waypoint and an entry lane waypoint. This virtual lane is potentially curvy, especially at intersections, hence it is important to understand the right way to represent and (re)-interpolate the curvy virtual lanes.
In sum, the goal of this project is to develop a principled approach to generate high-quality maps and corresponding interpolation techniques for autonomous on-road navigation. Implementation, evaluation and comparison will be performed.