Planning in information space for a quadrotor helicopter in a GPS-denied environment

Abstract

Unmanned Air Vehicles (UAVs) have thus far had limited success in flying autonomously indoors, with the exception of specially instrumented locations. In indoor environments, accurate global positioning information is unavailable, and the vehicle has to rely on onboard sensors to detect environmental features and infer its position. Given that a vehicle small enough to fly indoors can only carry a limited sensor payload, the vehicle's ability to localize itself varies across different environments, since different surroundings provide varying degrees of sensor information. Therefore, a vehicle that plans a path without regard to how well it can localize itself along that path runs the risk of becoming lost. My research focuses on how path-planning can be performed to minimize localization uncertainty, and works towards developing a motion-planning algorithm for a quadrotor helicopter. As a starting point, I apply the Belief Roadmap (BRM) algorithm, an information-theoretic extension of the Probabilistic Roadmap algorithm, incorporating sensing during the path-planning process. I make two theoretical contributions in this research. First, I extend the original BRM to use non-linear state inference via the Unscented Kalman Filter, providing better approximation of the non-linearities of laser sensing onboard the UAV. Second, I develop a sampling strategy for the BRM, minimizing the number of samples required to find a good path. Finally, I demonstrate the BRM path-planning algorithm on a quadrotor helicopter, navigating the vehicle autonomously in an indoor environment.