gold: gps and optic landing of drones a hybrid approach
TRANSCRIPT
GOLD: GPS and Optic Landing of DronesA Hybrid Approach
Stefania Maiman, Matt Schulman and Josh PearlsteinAdvisor: Jonathan M. Smith
Abstract— The successful use of drones for package deliveryoffers revolutionary cost savings for logistics providers. Today,the main barrier to drone delivery is the accuracy of dronesduring automated landings. GPS navigation accuracy is limitedto a four meter radius, so getting drones to accurately andinexpensively navigate to the last few meters of precision isa major obstacle. This project creates a new accurate andcost-effective solution for hyper-accurate drone landing. GOLDcombines GPS with optical navigation to iteratively recognizeand descend towards a target landing spot.
The system works by using the GPS navigation to send thedrone to a high altitude above the target. The target landingspot is denoted on the ground with a QR-code poster. Next,the drone iteratively photographs the landscape beneath it,processes the image onboard and gradually descends towardsthe ground based on the sublocation of the QR-code in eachphoto taken. This hybrid approach offers an inexpensive,scalable, and accurate drone delivery system.This project alsocreates a web interface to manage all drone flight, deliveryorders, and allocation of drones. For the proof of this concept,the website enables order fulfillment and allocation with adynamic simulation interface.
I. INTRODUCTION
In recent years, the discussion surrounding the use ofUAVs, unmanned aerial vehicles, or drones for commercialpurposes has been burgeoning. The commercialization ofdrones has the potential to greatly impact several industriesincluding agriculture, energy, utilities, real estate, infrastruc-ture, media and entertainment.
According to a recent report from Business Insider In-telligence, the market for commercial and civilian droneswill grow at a compound annual growth rate of 19 percentbetween 2015 and 2020, compared with 5 percent growth onthe military side[1]. Evidently, there is a huge potential inthe commercial drone industry especially when it comes tothe application of package delivery. Companies like Amazonand Google have already begun releasing videos showingtechnology capable of doing this with their projects AmazonPrime Air and Googles Project Wing.
While legislation continues to be a hindering factor in thedevelopment and use of commercial drones, there also needsto be a solution for the accuracy of delivery. Currently, UAVsrely on GPS technology to navigate from one location to thenext. However, currently GPS only offers an accuracy of 7.8meters with a 95 percent confidence interval, according to thelatest performance standards released by the United Statesgovernment[2]. While this accuracy can manage to get adrone from one remote location to the next, it is not sufficientwhen it comes to package delivery on a specific target
landing spot. A main barrier for drone package delivery to bematerialized is getting to that next level of precision. GOLDaims to create an innovative solution to this problem byoffering a cost-effective, portable, and scalable system forhyper-accurate drone delivery.
Multiple attempts are currently underway to attack thisproblem of getting to the next level of required precision[4]. A collection of technologies including beacons andother radio frequency identification devices are being used toengineer precise solutions. Still, the attempt to create a hyper-accurate landing solution that is affordable and scalable formillions of users has yet to be cracked; herein lies the focusof this senior design project.
II. APPROACH
A. Landing Algorithm
GOLD uses a hybrid approach to obtain accurate dronelanding by combining the power of GPS with optics to get tothat next level of precision. Ideally, a customer would makean order and then be sent a unique QR code to identifythat particular delivery. In a given time slot, the customerwould place the QR code on the ground near their residenceaway from any obstacles (trees, porches) that would preventthe QR code from being seen from above the ground, forexample, a backyard, a rooftop, a front lawn etc. The firststage of delivery is to send a drone to a specified GPSlocation. The next step is where the innovation comes in witha fine-tuned algorithm that allows for the drones autonomous,accurate landing. In this stage, the system would iterativelytake images from the drone of the ground below, processthem, and descend until it lands on the QR code on theground.
GOLD harnesses the power of GPS to get the drone froma starting location to the relative location for landing. Adrone is sent to a relative target location via GPS and hoversroughly 10 meters from the ground. Then, the algorithmtakes over for landing. This is how it works: the drone takesan image of the ground beneath it, processes it on-board torecognize where the QR code is relative to that image, andmoves in the direction of the quadrant the QR code is in.Since GPS has an accuracy of 7.8 meters, the QR code inthat location should be visible in an image taken from thatheight from the drone. After roughly 4-5 iterations of thisalgorithm, once the drone reaches a threshold of 0.5 metersabove the ground, it lands on the QR code.
B. Technical Stack
The crux of GOLDs stack is a Raspberry Pi attached tothe drone that does the optics analysis while the drone is inflight. The entire stack is visualized below:
Fig. 1. The stack.
The drone takes a picture of the ground beneath it withits camera. The drone then sends the imagen taken to theRaspberry Pi which is attached and connected to the drone.The Raspberry Pi is powered with a portable battery pack.The image is then processed in a C++ program written onthe Raspberry Pi that uses OpenCVs open-sourced imageprocessing library. The Raspberry Pi then comes up with arecommendation for how the drone should move based on thesub-location of the QR code in the image taken for verticaland horizontal movement. The Raspberry Pi sends this signalback to the drone, the drone moves, and the process repeats.Meanwhile, in parallel, there is a cell modem attached tothe Raspberry Pi that sends data to be analyzed through anssh tunnel to an AWS server that was set up. Data can beanalyzed and collected here. In addition, post requests areperiodically sent to a Ruby on Rails endpoint which canthen be used as a dashboard to monitor drones progress.That status dashboard can be viewed here: http : //drone−optics− app.herokuapp.com/dronemicrolanding.
III. RESULTS/ MEASUREMENT
After several stages of testing and refining the algorithm,GOLD was able to obtain an accuracy of 0.5 meters asopposed to the accuracy of 7.8 meters offered from GPS.
Once the final solution and algorithm was reached, theaccuracy was measured by taking measurements from thecenter of the QR code (target landing) and the exact locationof the drones landing. After several tests, it was found thaton average the drone landed within 0.5 meters of the target.
IV. ETHICS AND PRIVACY CONCERNS
GOLDs solution brings up a few legal/ethical concernsworth considering. There is already skepticism surroundingthe use of drones with high resoltuion cameras in publiclocations [5]. Due to the nature of GOLD, there are ethical
concerns when dealing with the use of such a camera in anarea where pedestrians may be captured in a photo. Whilecurrently it is legal to take photos from a drone in recreationalpurposes, there isnt a lot of specific legislation surroundingthe use of photos from a drone in commercial applications[3]. While the Federal Aviation Administration is in theprocess of releasing new legislation for commercial drones, itis worthwhile to come up with a solution that can eliminateany ethical ambiguity [3]. For example, for the future ofGOLD it might be possible to pre-process images and blurout any human faces before doing any processing of it.
Another concern similar to the one just mentioned dealswith the privacy concerns highlighted by the use of a cameranear residencies. While a customer that opts for deliveryusing a drone might be fine with images being processed oftheir property, the adjacent neighbors may not be. Dependingon the location of delivery, this could lead to future problemswhen using high resolution, real time images for dronedelivery. While ideally the images processed are only ofa small part of the ground or rooftop, the environment ofdrone delivery is so new it is difficult to predict what kindof locations/spaces drones will be delivering to. This concernwill also need to be addressed to prevent getting involved inany legal issues in the future.
V. DISCUSSION
A. Strengths
Our team is really excited about what we were able toaccomplish this semester. We developed a primitive landingsolution that offers a scalable and inexpensive autonomousdrone landing system in a generic fashion. While previoussolutions offered minimally accurate landings, GOLD wasable to achieve an unprecedented degree of accuracy. Wehope to move on to patent and commercialize our software.
A strength of our software is that it is completely portable.The GOLD system is not specific to any platform and cantherefore be attached to any drone. This allows for flexibilitywith its use and scalability in future applications.
B. Limitations
One of the limitations with what we achieved is batterylife. During our testing, we were limited by the Matrice100s poor battery life. The Matrice 100 only had enoughbattery for two or three tests before we would have torecharge it. This is definitely a limitation when consideringhow to commercialize this new technology, though it shouldbe noted that new drone technology is being developed withsuperior battery life such as the Matrice 600 which has sixtimes the battery life of the Matrice 100.
Furthermore, there are regulatory limitations when itcomes to the future of using drones for commercial purposes.Currently, the FAA is in the process of allowing the use ofcommercial drones, but there is still legislation that limitsthis use which varies from state to state [3].
C. Future Applications
Future directions of the project include verification ofdelivery, and outfitting our algorithm to fit special circum-stances like medicine delivery or urban delivery. The projectcould be modified to take a picture of the QR code, sendthe image of the package on the QR code to the recipientor create a cryptographic hash with the timestamp that couldonly be generated by the drone at the landing spot at theexact time. In this way it could provide an exact proof ofdelivery that no current delivery mechanism provides.
In addition, we could outfit the algorithm with specialcircumstances, for instance to land at a refugee camp withmuch needed medicine or to land on the top of a highrise building in manhattan to deliver packages. These futuredirections are not large departures from the current state ofthe algorithm and could vastly open up the possibilities withthe project.
APPENDIX
A link to a video displaying the drone autonomously land-ing using GOLD: https : //www.youtube.com/watch?v =IvXSFCkHBZo&feature = youtu.be
Fig. 2. Drone landing from the view of the drone.
Fig. 3. Image of the algorithm processing the field below it.
ACKNOWLEDGMENT
Jonathan M. SmithAni NenkovaDJI
REFERENCES
[1] ”THE DRONES REPORT: Market Forecasts, Regulatory Barriers, TopVendors, and Leading Commercial Applications.” Business Insider.2015. Web.
[2] ”GPS Accuracy.” GPS.gov. Web. 01 Apr. 2016.¡http://www.gps.gov/systems/gps/performance/accuracy/¿.
[3] ”Unmanned Aircraft Systems (UAS) Regulations & Policies.” FederalAviation Administration. N.p., n.d. Web. 01 Apr. 2016.
[4] B. Herisse, F. X. Russotto, T. Hamel and R. Mahony, ”Hoveringflight and vertical landing control of a VTOL Unmanned AerialVehicle using optical flow,” 2008 IEEE/RSJ International Conferenceon Intelligent Robots and Systems, Nice, 2008, pp. 801-806.
[5] Wild, Alex. ”The Ethics of Our Brave New Drone PhotographyWorld.” Scientific American Blog Network. N.p., 16 Feb. 2015. Web.01 Mar. 2016.