An Experiment to Assess Detection and Avoidance of Moving Obstacles from a UGV

摘要

In early FY07, the U.S. Army Research Laboratory (ARL) in cooperation with General Dynamics Robotics Systems conducted an experiment to assess detection and avoidance of moving obstacles from a UGV. The purpose of the study was to answer questions regarding human and vehicle detection and safe operations. The experiment exercised several detection algorithms being developed under the ARL program. Each algorithm had a designated computer shuttle on board the UGV, which drew sensor reports from among 3-D and 2-D LADAR, 2-D SICK, stereo and monocular video. The UGV was run on preplanned on-road routes, with the dynamic planner off. Over 176 runs, conditions were varied for UGV speed, moving obstacle speed and direction, MOUT verses Open terrain, and UGV route. Along the route, the UGV encountered a scripted array of "up to" 29 obstacles, some fixed (e.g. sticks, barrels, mannequins) and some moving (e.g. mannequins and vehicles). Algorithm detections were compared with these ground truth obstacles to determine response measures for the study, for example, the percentage detected and distance to ground truth at time of detection. This paper describes the experimental methodology, reports the approach to analysis, and shares some findings focused on human detection.