Previous Abstract Return to Session C4 Next Abstract

Session C4: Navigation Using Environmental Features

Satellite Image Informed Path Planning for Improved Visual Localization of Unmanned Ground Vehicles in GNSS-Denied Environments
Jordan Westphal, University of Illinois at Urbana-Champaign; Arturo Saucedo, Maribel DeLaTorre, and Ahmet Soylemezoglu, Engineer Research and Development Center
Location: Grand Ballroom IJ
Date/Time: Wednesday, Apr. 30, 4:00 p.m.

GNSS-denied and long-range unmanned ground vehicle (UGV) operations often occur in environments such as roads, grasslands, and deserts where a lack of visual features in the environment is an issue for visual localization. Common SLAM-based UGV navigation methods in GNSS-denied environments employ LiDAR, radar, or camera odometry, which drift significantly in visually uninformative environments. We present a path-planning approach utilizing a priori satellite imagery to maximize navigation in areas with high visual uniqueness. Offline, iterative template matching is performed on the satellite image to generate a map of visual uniqueness in the environment, utilized as a costmap to the global planner. The uniqueness map was tested on an autonomous UGV in a visually informative environment and a visually uninformative environment based on visual odometry dead-reckoning. Neutral performance was observed in the visually informative environment, while a 55% reduction in visual odometry drift from the ground truth was observed in the visually uninformative environment utilizing this approach.

Previous Abstract Return to Session C4 Next Abstract