IntroductionRelated WorkProblemApproachCoarse EvaluationFine EvaluationGoodness AssessmentExperimentsSetupResultsConclusionAIAA Infotech@Aerospace 20-22 April 2010, Altanta, GeorgiaOnline Assessment of Landing SitesSebastian Scherer∗and Lyle Chamberlain†and Sanjiv Singh‡The Robotics Institute, Carnegie Mellon University, Pittsburgh, PA 15213, USAAssessing a landing zone (LZ) reliably is essential for safe operation of vertical takeoffand landing (VTOL) aerial vehicles that land at unimproved locations. Currently anoperator has to rely on visual assessment to make an ap p roach decision; however. visualinformation from afar is insufficient to judge slope and detect small obstacles. Prior workhas modeled LZ quality based on plane fitting, which only partly represents the interactionbetween vehicle and ground.Our approach consists of a coarse evaluation based on slope a nd roughness criteria, a fineevaluation for skid contact, and body clearance of a location. We investigated whether theevaluation is correct for using terrain maps collected from a helicopter. This paper definesthe problem of evaluation, describes our incremental real-time algorithm, and discusses theeffectiveness of our approach.In results from urban and natural environments, we were able to successfully classify LZsfrom point cloud maps collected on a helicopter. The presented method enables detailedassessment of LZs without an landing approach, thereby improving safety. Still, themethod assumes low-noise point cloud data. We intend to increase robustness to outlierswhile still detecting small obstacles in future work.I. IntroductionLanding safely is an essential capability for all aerial vehicles; however, the majority of landings onunimproved terrain are performed by helicopters and other vertical takeoff and landing (VTOL) vehicles,increasing the risk of accidents. The situational awareness for teleoperation or onboard pilots is limited, soit is difficult to see small obstacles. A pilot can have diffic u lty judging the slope of terrain and identifyingobstacles at the landing site from a distance. Helicopters tend to tip over easily sin ce a large amount ofmass is at the top. Even very small obstacles such as railroad tracks can make an otherwise good landingsite unsuitable for landing.ANASAstudybyHarrisetal. analyzedrotorcraftaccidentsfrom1963-97andfoundthat36.19%ofthem were related to collision with objects, roll over, and hard landings.1Apartialcauseforaccidentsisalackofsituationalawarenessofthepilotregardingthesuitabilityoflandingsites. Reliabledetectionoflanding sites will increase the safety of operation by presenting vital information to the pilot before a landingapproach.Our algorithm evaluates the terrain using a combination of in a coarse and then a fine evaluation. Thecoarse evaluation find candidate areas applying a fast plane fit to measure the slope, roughness, and otherstatistics. The fine evaluation then evaluates promising areas by fitting a 3D model of the helicopter andlanding gear to a triangulated surface of the environment.The main contributions of this paper are• an analysis of the problem of landing site evaluation.• an incremental model-based method to calculate evaluate landing sites for VTOL vehicles.• results based on ladar sensor data that show the landing sites found in real environments.∗Graduate student, Field Robotics Center, [email protected]†Research Engineer, Field Robotics Center, [email protected]‡Research Professor, Field Robotics Center, [email protected] Institute of Aeronautics and Astronautics Paper AIAA 2010This paper first reviews related work, details our approach, and then presents results for the landing siteevaluation algorithm.Where is a good landing zone?Point CloudOverhead ImageFigure 1: The problem addressed in this paper is to find good landing zones for rotary wing vehicles. Onthe left is an overhead image for reference. The in put to the algorithm is an incrementally updated pointcloud that is evaluated to find suitable landing zones.II. Related WorkThere has been some prior work on landing and landing site selection. From a control perspective theproblem has been studied by Sprinkle2to determine if a trajectory is still feasible to land. Saripalli et al.3, 4have landed on a moving target that was known to be a good landing site. Barber et al.5used optical flowbased control to control a fixed wing vehicle to land.Vision has been a more popular sensor modality for landing site evaluation because the sensor is lightweight.De Wagter and Muld e r6describ e a system that uses vision for control, terrain reconstruction, and tracking.AcameraisusedtoestimatetheheightabovegroundforlandinginYuetal.,7and similarly in Meijas etal.8asinglecameraisusedtodetectandavoidpowerlinesduringlanding.Using a stereo camera, Hintze9developed an algorithm to land in unknown environments. Bosch et al.prop ose an algorithm for monocular images that is based on detecting nonplanar regions to distinguish land-ing sites from non-landing sites.10Another popular approach is to use structure from motion to reconstructthe motion of the camera and to track sparse points that allow recon s tru c tion of the environment and planefitting. This method as been applied to spacecraft by Johnson et al.11and to rotor craft by Templeton etal.12Ladar based landin g site evaluation has not been studied very deeply except as su gge s ted in Serrano etal.13where a framework for multiple sources like ladar, radar, and cameras is proposed to infer the suitabilityof a landing site. For ground vehicles ladar based classification is popu lar because of its reliability and directrange measurement. It has been used in conjunction with aerial vehicle data for ground vehicles by Sofmanet al.14and by Hebert & Vandapel.15Our work in landing site evaluation uses ladar data and some of the proposed te chniques in related worksuch as plane fitting and roughness. However, our method goes beyond pure plane fitting to actually fit amodel of the helicopter geometry to a triangulation of the environment.III. ProblemThe suitability of a landing zone depends to a large degree on the characteristics of the aircraft th at will belanding on it. We consider larger vehicles such as manned helicopters, however the problem scales to smalleraircraft. Landing of a h elicop ter can roughly b e separated into two independent phases: an approach, andagroundcontact.