Prototype Sense-and-Avoid System for UAVs - The Robotics Institute Carnegie Mellon University
Home/Prototype Sense-and-Avoid System for UAVs

Prototype Sense-and-Avoid System for UAVs

Christopher M. Geyer, Debadeepta Dey, and Sanjiv Singh
Tech. Report, CMU-RI-TR-09-09, Robotics Institute, Carnegie Mellon University, May, 2009
View Publication


In this technical report we describe our efforts towards a field deployable Sense and Avoid system for Unmanned Aerial Vehicles (UAVs) which uses passive vision as the main sensing modality. The ability of UAVs to operate safely in the presence of other air traffic is critical towards acceptance of UAVs in civilian and military airspace. This will allow UAVs to be used to their fullest potential. In order to operate freely in the presence of manned airborne traffic UAVs must demonstrate a Sense and Avoid capability that meets or exceeds that of an equivalent human pilot. Furthermore this capability should be achieved without the use of cooperative communication with other aircraft or prior knowledge of other aircrafts’ flight plans. We describe our collision avoidance algorithm and software-in-the- loop testing, vision based detection method with a 98% detection rate out to a range of 4.5 miles which exceeds the FAA regulation of 3 statute miles. A field deployable Sense and Avoid system must be able to operate with consistent performance across a variety of atmospheric conditions including cloud, fog and haze of various degrees that can occur under conditions commonly described as Visual Meteorological Conditions (VMC). In order to model the effect of all these conditions on the performance of the detection system we developed an atmospheric image formation modelling system that takes as inputs weather conditions and can predict the appearance of the image of an aircraft of given geometry at various ranges. This predictive model also determines the minimum resolution needed to guarantee the required detection performance. Passive vision provides the bearing to the intruding aircraft. Range estimation is only possible by executing additional maneuvers which causes significant mission interference. We investigate the feasibilty of a flash lidar system that can be used as a confirming sensor to further reduce the false positive rate and provide range of the intruding aircraft.


author = {Christopher M. Geyer and Debadeepta Dey and Sanjiv Singh},
title = {Prototype Sense-and-Avoid System for UAVs},
year = {2009},
month = {May},
institution = {Carnegie Mellon University},
address = {Pittsburgh, PA},
number = {CMU-RI-TR-09-09},
keywords = {Sense and Avoid, UAVs, Computer Vision},