DARPA LAGR: Learning Applied to Ground Robots

DOD AFRL award # FA8650-07-C-7702
Award amount: 770,767.00
Award duration: 10/18/2006 - 7/18/2008

Investigator

Gregory Z. Grudic
I. Jane Mulligan

Abstract

The goal of this project is to apply Machine Learning and Computer Vision techniques to the open problem of navigating between two GPS weigh points, which are approximately 300 meters apart, in unstructured outdoor environments. The LAGR robot (Figure 1) is put into an environment it has never seen before, and is expected to intelligently find its way to the goal position. The LAGR program consists of 8 teams (GaTech, Netscale (NYU), SRI, NIST, API, JPL, UPenn, CU), all with identical robots and all competing at Government run competitions on a monthly basis.

Figure 1: The LAGR Robot

A key component of this project is far field navigation. The robot has a pair of stereo cameras that allow it to detect obstacles in the near field (less than 12 meters). We use Machine Learning and Computer Vision techniques to extend this near field information to navigation decisions made up to the 100 meters ahead of the robot.