Real-Time, Multi-Perspective Perception for Unmanned Ground Vehicles

The most challenging technical problems facing successful autonomous UGV operation in off-road environments are reliable sensing and perception. In this paper, we describe our progress over the last year toward solving these problems in Phase II of DARPA’s PerceptOR program. We have developed a perception system that combines laser, camera, and proprioceptive sensing elements on both ground and air platforms to detect and avoid obstacles in natural terrain environments. The perception system has been rigorously tested in a variety of environments and has improved over time as problems have been identified and systematically solved. The paper describes the perception system and the autonomous vehicles, presents results from some experiments, and summarizes the current capabilities and limitations.