Localizing Ground Penetrating RADAR: A Step Towards Robust Autonomous Ground Vehicle Localization
Abstract
Autonomous ground vehicles navigating on road networks require robust and accurate localization overlong term operation and in a wide range of adverse weather and environmental conditions. GPS/INS solutions, which areby themselves insufficient to maintain a vehicle within a lane, can fail due to significant radio frequency noise or jamming,tall buildings, trees, and other blockage or multipath scenarios. LIDAR and camera map-based vehicle localization can fail when the optical features change or become obscured, such as with snow, dust, or changes occurring on the surface of gravel or dirt roads. The optical surfaces themselves are also easily susceptible to damage from small rocks or debris. Localizing Ground Penetrating RADAR (LGPR) is a new mode of a-priori map-based vehicle localization designed tocomplement existing approaches with a low sensitivity to the failure modes of LIDAR, camera, and GPS/INS sensors due to its low frequency RF energy, which couples deep into the ground. Most of the sub-surface features it detects are inherently stable over time. Many areas of research, discussed in this paper, remain to prove out general use of the concept. We have developed a novel low-profile ultra-low power LGPR system and demonstrated real-time operation underneath a passenger vehicle. A correlation maximizing optimization technique was developed to allow real-timelocalization at 125 Hz. Here we present the detailed design and results from highway testing in which we use a simple heuristic for fusing the LGPR estimates with a GPS/INS system. Cross-track localization accuracies of 4.3 cm RMS relative to a truth RTK GPS/INS unit at speeds up to 100km/h (60mph) are demonstrated. These results, if generalizable, offer a widely scalable real-time localization method with cross-track accuracy as good as or better than current localization methods.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 27, 2015
- Accession Number
- AD1034980
Entities
People
- Beijia Shang
- Byron M. Stanley
- Jeffrey C. Koechling
- Matthew T. Cornick
Organizations
- MIT Lincoln Laboratory