FCA US to invest $74.7M in Trenton Engine Complex for new 4-cylinder
Volvo Cars to launch UK’s largest autonomous driving trial

SwRI to showcase Ranger precision localization technology for automated driving; non-GPS system with 2cm precision

Southwest Research Institute (SwRI) will showcase its award-winning Ranger precision localization solution at the AUVSI XPONENTIAL 2016 conference and trade show in New Orleans 2-5 May.

Ranger is a patented approach to vehicle localization that enables precise navigation for automated vehicles using commercially available hardware in combination with SwRI algorithms. The latest Ranger kit can be used for automated driving, valet parking in garages and structures, freight distribution, and docking of buses and large trucks.

We have made this technology smaller, faster, and more robust for real-world use at a relatively low cost.

—Dr. Kristopher Kozak, leader of Ranger’s development

Ranger is a hardware/software solution; it uses a ground-facing camera and localization algorithms to provide precise position and orientation measurements. Ranger images the unique “fingerprint” of road surfaces by matching thousands of distinguishing ground features, such as aggregate, cracks and road markings, to corresponding features collected and stored in a map.


Using a ground-facing camera and algorithms, Ranger allows for precise automated driving within 2 centimeters. Top. A view from the underside of a research vehicle equipped with the Southwest Research Institute’s Ranger localization solution for autonomous vehicles. Bottom. Side view illustration of placement. Source: SwRI. Click to enlarge.

To operate, Ranger takes advantage of the fact that individual patches of ground are uniquely identifiable from the distribution of visible features on the surface. For most types of pavements (asphalt and concrete, for instance), these features frequently correspond to distinct visible features such as stains, cracks, tar patches, exposed aggregate, etc., but other less distinct features and textures can also uniquely distinguish ground patches. The uniqueness of patches of ground serves as the basis for Ranger localization – matching a current view of the ground to a previous observation of the same patch of ground provides very precise position information. To create a full localization system a map of ground images must be built. And to do this, images of the ground are aggregated into a map, which can be queried by location.

A Ranger map conceptually consists of images that are registered to a global, fixed frame. If the registration of the frames is accurate, then Ranger measurements are both precise and accurate. To ensure local and global consistency of the map, a global bundle adjustment-like optimization method was developed and used as a final refinement step to the raw map. This optimization balances factors such as the original camera position measurement (captured during map data collection from a GNSS receiver) with the relative position of sequential frames based on the kinematics of the data collection vehicle with relative position constraints between overlapping image frames (e.g. single segment loop closures, inter-segment overlaps, etc.) based on image matches.

Once a map is built, localization is performed by matching a live image frame to an image frame stored in the map. The matching process is based on a conventional image feature matching approach that has been extensively refined and optimized for this application. The basic approach is as follows:
1. Detect image features (Detect Keypoints, Extract Feature Descriptors).
2. Match features in live image to candidate map frame.
3. Apply geometric constraint (rigid transformation) to matched features using RANSAC techniques – the specific rigid transform, which is computed as part of this step, gives the precise relative rotation and translation between the live image and the map image.
4. If a sufficient number of matches (inliers) satisfy the rigid transformation constraint, then the live frame positively matches the map frame.

—Kristopher Kozak and Marc Alban (2016)

Ranger is a novel approach to localization for automated driving that performs well in all illumination conditions and typical adverse weather conditions, including rain and fog.

Other sensor-based localization approaches may be more common, but Ranger’s camera-based approach with controlled illumination is unique and extremely accurate. This allows for precise automated driving within 2 centimeters, similar to the most accurate (and most expensive) GPS systems. Ranger, however, can operate in areas or environments where GPS has poor performance or fails completely, such as in tunnels or underground.

GPS is ubiquitous, everybody has GPS on their phones, but it’s not always as accurate as you need it to be for automated vehicle localization. Ranger is a low-cost, high-precision localization system that overcomes a lot of problems affecting GPS systems.

—Kristopher Kozak

In April, Kozak’s team earned the Walter Fried Award for Best Paper at the IEEE/ION PLANS 2016 conference in Savannah, Ga., for technical work documenting Ranger innovations. In 2015, R&D Magazine named Ranger as one of the 100 most significant innovations of the year.




Amazing if it can be done through snow, ice, rain, pot holes, patches, dirty camera lenses etc.


@Harvey, Exactly: it won't work with snow or rain or ice.
Rain on the lens (or splashed up rain) will also mess up the image.
+ How big are the maps ?
They must be huge - probably 1mm resolution.
Do the math on the size of them for say 2m wide x 100km long.

Maybe they only plan to use it in the applications given:
"valet parking in garages and structures, freight distribution, and docking of buses and large trucks."
These would have small maps. The broader application of "automated driving" would have enormous map requirements.

The comments to this entry are closed.