PeRL STUDIES AUTONOMOUS NAVIGATION & MAPPING FOR MOBILE ROBOTS IN A PRIORI UNKNOWN ENVIRONMENTS.
PeRL studies autonomous navigation and mapping for mobile robots in a priori unknown environments.
Next Generation Vehicle (NGV)
The Next Generation Vehicle Project was a collaboration between PeRL, APRIL, and Ford Motor Company. The project consisted of several Ford Fusion Hybrid automated research vehicles that are being used to make progress on automated driving and other advanced technologies.
Our Ford Fusions were equipped with four Velodyne HDL-32E laser scanners, an Applanix POS-LV 420 inertial navigation system, and a single forward-looking Point Grey Flea3 camera.
Visual Localization within LIDAR Maps: We demonstrate our vehicle localizing with a single monocular camera within a 3D prior ground-map, generated by a survey vehicle equipped with 3D LIDAR scanners. To do so, we exploit a graphics processing unit to generate several synthetic views of our belief environment. We then seek to maximize the normalized mutual information between our real camera measurements and these synthetic views. With this methodology, we are able to achieve localization accuracy similar to LIDAR-only configurations.
Generic Linear Constraint Node Removal: The generic linear constraint (GLC) framework provides a method to remove nodes from SLAM graphs. This can be used for graph maintenance and to reduce computational complexity during long term SLAM. GLC produces a new set of factors over the elimination clique given only the existing factors as input. The new factors can represent true dense marginalization or a sparse approximation of marginalization. GLC works in graphs with less-than-full DOF constraints (e.g., bearing-only, range-only), and avoids inconsistency found in methods based on measurement composition. Software is available here: GLC Software.