Covariance estimation for GPS-lidar sensor fusion for uavs

Outdoor applications for small-scale Unmanned Aerial Vehicles (UAVs) commonly rely on Global Positioning System (GPS) receivers for continuous and accurate position estimates. However, in urban areas GPS satellite signals might be reflected or blocked by buildings, resulting in multipath or non-line-of-sight (NLOS) errors. In such cases, additional onboard sensors such as Light Detection and Ranging (LiDAR) are desirable. Kalman Filtering and its variations are commonly used to fuse GPS and LiDAR measurements. However, it is important, yet challenging, to accurately characterize the error covariance of the sensor measurements. In this paper, we propose a GPS-LiDAR fusion technique with a novel method for efficiently modeling the position error covariance based on LiDAR point clouds. We model the covariance as a function features distributed in the point cloud. We use the LiDAR point clouds in two ways: To estimate incremental motion by matching consecutive point clouds; and, to estimate global pose by matching with a 3-dimensional (3D) city model. For GPS measurements, we use the 3D city model to eliminate NLOS satellites and model the measurement covariance based on the received signal-To-noise-ratio (SNR) values. Finally, all the above measurements and error covariance matrices are input to an Unscented Kalman Filter (UKF), which estimates the globally referenced pose of the UAV. To validate our algorithm, we conduct UAV experiments in GPS-challenged urban environments on the University of Illinois at Urbana-Champaign campus.