Journey to Site Scan: Lessons Learned Building Drones for Construction

NICK SPEAL: Hello and good morning, everybody. My name is Nick Speal and I'm a software engineer at 3DR. And at 3DR we make a product called Site Scan, which involves a drone that can fly over your job site and collect photos, and take those photos and upload them to the cloud, where they can be stitched together. And then you can run analysis and visualize that data from a web browser. So today I'm going to speak a little bit about Site Scan from an engineer's perspective. I'm going to go through some of what I found to be the more interesting technical details about how we built this product and how we got to where we are today.

Then I'm excited to announce a new feature that we're launching this week at Autodesk University, which involves integrating Site Scan with the Forge API. And finally, I'm going to reflect on some of the values that we have that got us here today. But first I'm going to teach you how to fly a drone. So can I get a show of hands, who here has a commercial pilot's license? Couple of you. The good thing is you don't need that. All you need is an iPad.

So you can select area survey. You draw a rectangle around your drop site and you want to specify the altitude that the drone flies. After that, all you have to do is press one button. The system verifies that everything is ready to go and you swipe to take off. It's really that simple. But it wasn't always that simple. When I joined 3DR three years ago, one of these appeared on my desk. And this was a drone called the Do It Yourself Quad Kit, and it was a product of 3DR sold at the time.

And the idea was you have all sorts of different parts and you have an IKEA-like instruction guide that you use to put it all together. And so after you put it together, you fly it around until you crash it, and then you put it together again and that's the fun of it. So we've come a long way since those hobby days making drones much more easy to use and much more reliable. So this is a little video that shows where we are with Site Scan today.

[MUSIC PLAYING]

So this video showcased a lot of the different features that we have for making it easy to fly a drone and what you can do with the data once after the drone has flown and collected the photos. Today I'm going to talk through some of the main pieces of functionality that goes into making this possible. So first there's autonomy. So what makes the drone easy to use and collect this data. Accuracy, how do we make the data accurate and how accurate is it after all? What do you do with accurate data? Well measurement and quality control.

So to consider autonomy, let's look at this preflight checklist that we saw in the video. It's very simple. All you have to do is swipe to take off. But sometimes behind this really simple user experience is more complicated under the hood. The system is performing lots of different checks to make sure that everything is safe to fly every time you're out in the field. Consider for example the sensors. The compass for example. Those three compasses actually onboard a drone for redundancy. And one of the checks that happens before you take off is we want to make sure that the values on all three compasses are the same because if they're not perhaps one of the compasses could be damaged or miscalibrated.

And so before letting you take off, we make sure all the values are the same on all the compasses. Well, one of the problems with this that we found out about is sometimes people like to fly their drones from a metal surface, especially on a construction site. So maybe you want to take off from the bed of a pickup truck, or a steel plate, or concrete with embedded rebar you can't even see. Well there is this one compass that's on the very tip of the leg of the drone you can see here. And that makes it further away so it's isolated from the motors or from any of the other or other electronics onboard, but that makes it closer to the metal surface if you're trying to take off from a metal surface. And so what we've seen is you actually get a different value reading on the compass on the compass that's closer to the metal or the ones that are further away.

So in order to get around that, what we did is we implemented a latching check. So we said these campuses need to be consistent for three consecutive seconds sampling 10 times a second, 30 times in a row. If you get the exact same values and all of them you know that your sensors are healthy. After that, if you're getting inconsistent values it's because the user brought the compasses or the drone closer to a metal surface and you know that that anomaly will clear after takeoff. So this is one feature that we have that enables safety and autonomy and makes it really easy to fly drones.

Another one that is very early back in drone development was the autonomous battery failsafe. So this is the simple idea that when the drone gets low on battery, it can fly by itself back to the GPS location where it took off from and land and turn off its motors. Well one consideration is if you've got the drone here and the home location here, what happens if there's a tree in between? Well you don't want your drone to end up in the tree. And so what it does is it'll fly up to a high safe altitude, say 25 meters, then fly home and then descend, and that way you're avoiding obstacles in between.

Another consideration here is how far is the drone away from home? If you say, OK, low battery means 20%. Once the drone is at 20% it should go back to the home location. Well if the drone is half a mile away, and yeah they can fly that far, then 20% battery is not enough to make it home. Whereas if the drone is right close to the location then you're leaving a lot of battery on the table shortening the battery life by returning home too early. So what we do is we measure the amount of current that the drone is consuming and we estimate how much time it will take for the drone to fly up to the safe altitude, back to the home location, and then land. So based on this time estimate and the amount of current that it's consuming, we can estimate how much energy it's going to take to get home and make sure we leave at the point so that the drone lands with a buffer of 10% battery remaining.

This is another feature that makes it one less thing that the operator needs to worry about so you know that the drone's autonomously going to come home before it runs out of battery, no matter where it is. So with this autonomy, you've got a drone that can fly around safely all by itself, what do you do with that? Well what our drones do is take photos of your construction site automatically and all over the site at the precise locations to get full coverage. With this data, it uploads it into the cloud and stitches together 2D and 3D data products we call them. So for example a 2D map, an orthomosaic, a bird's eye view of your whole site, or a 3D model, for example a point cloud. Well how accurate is this is this data?

Well you can take this 2D map and you can zoom all the way in and you can zoom right in. You can find that particular spot on site, say X marks the spot. And in a web browser you can just click it and see its exact coordinates. Alternatively, you could take the most expensive GPS you could buy and walk out on site and survey that spot. And you're going to get the same value to within 1/10 of a foot. So you want to take a measurement. You take a big tape measure out and you can measure the distance from point A to point B, you're going to get an accurate value. Or you can do the same thing in a web browser and get the same reading to within 1,500ths of a foot.

So how do we get this data to be so accurate? Well it starts with the GPS on the drone. So the drone has a GPS onboard of course and it uses this for navigation to know where it's supposed to be able to go. But it also saves a log of every time it takes a photo, this is my exact location of where the drone was when it took that photo. Then when the photos are uploaded for processing, the photogrammetry engine uses this GPS coordinates of where the drone was to project down what are the coordinates on the ground. So this is a visualization of what the photogrammetry engine is doing.

Say all of the blue dots represent the drone's location when it took all those photos. What it actually does is it has overlap, maybe five or six photos of every point on the site. And it uses that to adjust those locations and wiggle them and make sure everything aligns and get you a more accurate representation of where each point is. But unfortunately that's not enough to get you the level of accuracy that I just talked about. For example, one thing you might not expect is that the accuracy of GPS depends on the weather. It depends on the day and just the content of the atmosphere on any given day.

And since this will affect the accuracy of all of your photos, it'll shift them all one way or another by a couple of feet. So in order to work around that, we use what we call ground control points. So this is a known location on site, X marks the spot, and you survey that location with a corrected GPS. A corrected GPS will get measurements from a known base station somewhere nearby and can use those measurements to correct for any sort of deviations in the atmosphere and precisely know the location of one piont to a very high degree of accuracy.

And so we have a workflow where you can go and you can click the photo and say, OK, this pixel is that my known ground control point. And you do this for maybe four or five points around your site, and then the photogrammetry engine can constrain those points to be exactly where you said they'd be. And if you know this point is accurate and this point is accurate, you can use checkpoints to note and verify that the points in between are also accurate.

So with this accurate data easily accessible in a web browser, what can you do with that? Well, the first thing is measurement. So it's easy to drop a marker and get the 3D XYZ coordinates of a point. You can measure the distance from point A to point B or the area within a region of interest. But one of the more interesting features from a technical point of view is volume measurements. So say you have some stockpiles and you want to know how much earth there is. You can simply draw a polygon surrounding your stockpile and quickly get the cut and fill measurements for the volume within that region.

Well how does this work? What we have is a massive array of all of the different heights of every pixel on site, so to calculate the volume, we take the area of that pixel, you know its length times its width, and you multiply that by that height. But it's not quite that simple, because say a particular coordinate on site is at 5,000 feet above sea level. Well you don't really want that volume of that distance all the way down to sea level. You're not going to go dig all the way down to sea level even though there's a bunch of volume of earth there. What you're really probably interested in is the distance between the surface on top of the pile to the ground underneath it.

But how do we measure that? We don't have a direct measurement of that. The drone can't see the ground underneath the pile. It's obscured by the pile. And so what we do is as the user draws a polygon around the pile, we use the altitude and the height of these points around the pile and we create a surface shown here in yellow. This is a 3D surface, for example it works on a sloped plane, and we create a surface that goes in between and connects this perimeter that the user draws. From that, we can find the distance between this base plane and the surface above.

Now as a user, if you see this pile is 149 yards, it's not always you know obvious what are you measuring? Is this interpolation working as I'd expect? And so we built this volume visualizer where the user can see exactly what they've selected and make sure that the volume that they're collecting and measuring is what they had in mind. One of the things that sort of amazed me when I first started working on this as a web developer is that all of this is actually happening in the browser on a user's computer in real time. So this isn't being outsourced off to some server for processing. This is happening live right on your computer.

And so in order to make this happen we have to consider performance. And as it turns out doing these volume calculations really is quite quick. What takes a while is rendering this mesh that you can see here on the right. And so this is a mesh connecting all of the different points surveyed here on the surface of the pile. And for a small pile like this, it may only take a second or two, but for a large region, for example, if you select the whole site, this can take up to 20 seconds. Now you might say, OK, I'm patient, I can wait 20 seconds. But consider that our web application is refreshing at 60 Hertz. So every 60th of a second you're seeing a new image representing our web app, and this allows us to be able to provide animations when you click and drag, it's responsive.

And so if one operation is taking 20 seconds, then there's going to be a 20 second lag between you clicking a button and then getting any sort of feedback. And so that's not acceptable for responsive web application. So what we did to get around this is we take this massive mesh millions of points here within the bounding box and we divide it into 10,000 smaller meshes. And we render each one at a time. Then after each mini mesh is rendered, we return control to the browser to be able to update. And so what you end up with there is this animation that you can see on the right the sort of scanning animation as we render each slice of the mesh one at a time. And in between each of these slices we're able to update the browser and so you can see the map on the left is able to smoothly be dragged around while this is being rendered.

So that's volume measurements. What else can we do with this accurate data that's autonomousally collected from a drone? Another is quality control. So imagine you have a bird's eye view of your site as it stands today. That's valuable. How can you use that to make sure that you've got quality control on site? Well say you have the plan data for what your project is supposed to look like in a geospatial format, maybe Google Earth's KML or as a re-shape file. You can just drag that right into Site Scan and overlay it right on top. From that you can zoom in and you can see any examples of where the asbuilt and as planned don't exactly line up.

One problem however with this feature that we started to notice is that not everybody has their data in a geospatial format like a KML shape. Instead, what people often have is a PDF or can easily convert to that. And so to make this easier to use, we wanted to support PDF overlays on top of your site. Now a problem is how a PDF isn't a geospatial format. So how do we make this work? Well, we built a workflow where you tag two points on your map and and those same two points on your PDF. And then from that we can measure the scale in rotation such that they can line up exactly.

Now what we do is we actually render the map as a canvas and then the PPF on a canvas overlaid on top of that with an opacity. And we use CSS to transform that PDF and stretch it, scale it, or rotate it such that it stays in sync with the map. Every time the user clicks and drags, we had an event and we can update that CSS and keep the overlay in sync.

These are some of the features that makes Site Scan data very powerful to be able to use and do analysis with. Now I'm excited to talk about one of our new features that we're launching this week at Autodesk University and makes it easier to share data across your team. And this uses the the Forge API to integrate with BIM 360.

So there are a couple of features associated with this. One is first of all just connecting your Site Scan account to your Autodesk account such that administrators can control access in one BIM 360 admin page and make sure that their data is being commissioned as they expect. This file import, so you've got files in BIM 360 Docs, you can overlay them as a PDF right on the map on Site Scan. And if a new version of that PDF gets released on BIM 360, it's automatically updated in Site Scan so you know nobody's looking at old data, making decisions with old data. You can export your Site Scan files so your 3D point cloud for example, you can send that right into BIM 360 for use in other programs like maybe Civil 3D and share it across your team.

And if you discover an issue in Site Scan, it's very easy to just click a button and report an issue so that people in the field can correct it. So to talk about this feature and really understand how is it useful on sites, I'd like to invite to the stage one of our customers who's been really on the forefront of integrating with new technologies. And he's been working together with us in terms of developing this feature and making sure it's just right. So ladies and gentlemen, please welcome Andy Leap of PARIC.

ANDY LEAP: Accurate stage. You going to be OK? You guys hear me OK? All right. So that is a grand entry. Thank you Nick. OK so just to give you kind of a little overview. So how many people in here are using drones in a workload today? So how many people are scanning with a drone today? OK, all right. So just a little bit about PARIC. We're a St. Louis based construction company. We work on all kinds of projects, vertical projects, in various capacities. We're a growing firm. We've grown quite a bit here in the last few years. And so each year seems to be getting better and better so long as the economy holds out, I think that's great for all of us. There's plenty of work to go around.

And so with that it's kind of lead us to cross paths with 3DR and what they were doing was we needed to be able to start really capturing the day to day workflows that we were doing. And so kind of move on past all the-- OK. We started using drones a couple of years ago. And so what we were seeing was we could get all kinds of imagery, all kinds of videos, but we didn't have any real data we could overlay into our workflows. And what we really were looking for initially was pictometry so that we could use it in pre-construction phase of the work.

And so what that meant for us was we wanted to go out and scan a site, we couldn't rely on like a Google Earth image necessarily because those maps get out of date pretty quickly. So we wanted to go out in the middle of literally the middle of nowhere, some of our job sites are pretty rural. And so we wanted to get really good data so we'd understand the topo of the site, roads, access, you know context around buildings, that kind of thing. And so for us to be able to put that put drone on a plane, take it with us, go capture the information and bring it right back with something that shortened our workflow so we actually had accurate information.

And so from that point we can build some simulations. And [INAUDIBLE] we're going.

So what that allows us to do is create some pretty crafty simulations where we take a model-- whether it's from Revit, or CAD, or SketchUp-- and overlay it in real context. So we can start to study things like logistics-- how are we going to access this site, how are we going to do our excavation, how are we going to set up the job site.

We do a lot of large campus projects where there's people that are actively going to be using that site-- driving, walking. In the middle of winter, they need to get from one building to the next. So we're able to create these really crafty simulations that show this in as much realism as we can depict to a potential client, an owner, and also to the project team.

Everybody sees something different when they look at it. A project manager or somebody who's going to be delivering materials see sites, and the access points, and that kind of thing. But the owner goes, wait a minute, you're going to impact where I'm parking my staff, or you're going to impact how people visit my campus. And so all of a sudden they see a different perspective that they don't see on a set of 2-D plans.

And this is where we reached out and started working with 3DR about the information that we were getting was great, but we needed formats that would really work to support this type of a workflow, and do it not only quickly but also allow us to iterate. So we may go out and scan. And as we're going to the project, things on-site are changing. So we maybe go back and update this logistics simulation. So that allows us to go through this process repetitively.

So where the conversation got really interesting was we sat down and mapped out all the software we use. Has anybody used this much software in a project besides me? So you start to get a feel for how much software it takes to create something like that.

And it's not just Revit. It's not just a simulation tool. It literally is a variety of tools that we were using.

Now, the thing about PARIC is we're an enterprise customer within 360. So we want to really use every piece of that tool that we. So we want to leverage the data across our entire team and make sure they all are using the same information all the time.

So we started talking with 3DR. And, fortunately, they decided they would fly in and spend a few days to understand or workflows more face-to-face. So, took them out to the job site, talked about this. And we threw this up on the screen and obviously they were like, wow, that is a real mess.

But we had a vision of what we wanted to accomplish. And so what's really exciting about the Forge and the entire platform is that it's opening doors for us to have conversations with developers like this that-- we're not coders, we're not computer scientists. We don't know the first thing about creating these types of hooks to make this information work. But they do.

And the downside to their work, though-- or not the downside, but the piece that they're missing, is understanding the workflow and the things that we actually need. So how do we work together to create this content?

So we mapped this out and talked about what we needed. And they went back and started working on this. And so within a very short amount of time they built a prototype. And so that allowed us to create the simulation I showed you. But it also allows us to do some other things.

So early on we were doing production tracking, which Nick showed you. And this has been a hugely powerful tool in just being able to animate from week to week what has changed on a job site-- tracking concrete pours, how much earthwork has been moved around, how much utility lines have been placed, how much steel has been erected. All that stuff matters on an active job site when you're trying to plan work and who's going to be showing up on the site the next week.

So that was one step. That's great and that's very helpful. But how do we actually share this information?

So what we wanted to be able to do next was be able to import and export content. So Nick talked about this a little bit earlier, but we have actual design files that are part of an active project. But they're all sitting out in our BIM 360 library in docs. So how do we create this workflow where we can get the active updated drawing that may change from week to week in that early process of a project and bring it in?

And so we had to give up a little bit on our permissions so they could get into our job or our projects. But there's quite a bit of comfort there in knowing that everything's still locked down and all the permissions are there.

So we worked through these workflows. They did a great job of being able to create an overlay process. And it's very simple to use. And so we've already started using this on our projects. And the team really loves the access. Because now they can really track what's going on around excavations, what's going on with steel erection, and how they're sequencing that work.

But taking that next step further, with BIM 360 we have an entire system built around issue management. It's a database like any other tool. But for us, it's our day to day-- what's going on on the project, what needs to be fixed, what needs to be addressed on a job site. And so to create a workflow where we can leverage the issue tools inside of BIM 360 is yet another thing that they brought to the table.

So now we can actually identify an issue, whether just visually or from measuring, and do it within sight of the Site Scan tool set. And these are pushing to BIM 360. So again, a way that we can leverage our existing tool set, but by doing it through a slightly different frame.

And the measuring of this stuff is actually amazingly accurate. In the construction world you're never, ever, ever supposed to measure a paper drawing set. How many people have ever heard that? Never measure paper drawings.

So we're not looking for 100% accuracy on it. But it is amazingly accurate, especially when we're trying to figure out just the basic things on a site around just moving earthwork, or piles of rock around, or how much concrete has been poured in a day. And we can fill out areas and distances pretty quickly.

So the next is exporting it. So is it playing? Hopefully it's playing.

So the thing we're going to do is we've got all this amazing data in Site Scan. And this was actually the crux of our initial conversation was you've got all this amazing data inside of Site Scan-- how do we get it out and get it into our tool set?

And so the challenge that we had was the files are huge. You may scan a site, and pull out a Pictometry map, and it's anywhere from 50 megs to 450 megs. It just depends. And when you're doing this repetitively, you don't want to have to download, grab the information on your local drive, and then push it right back up into BIM 360. It's a time vampire. So how do we get around that?

So this is where the initial conversation started of how do we build a bridge that can allow us to leverage that data with the entire design team and the build team at the same time. So 3DR built this link where it's really easy. It's basically a simple click. You click the Export box and you just click the link. It automatically knows. Once you've set up that path once, it will push the data to the same folder every time. So with that, I'll turn it back over to you, Nick.

[APPLAUSE]

NICK SPEAL: Thanks, Andy.

ANDY LEAP: Yep.

NICK SPEAL: You know, we work day in and day out building software for our customers. And so to see it in action in the broader context of how it's used on-site is really rewarding. So, thank you.

So these are some of the functionalities that Andy talked about it in terms of integrating data with the larger team through BIM 360. And I want to take a few minutes to dive into some of the technical details about how did we build this with the hope that some of you might be thinking about building it-- something like that-- and we might be able to learn something from that.

Now a lot of it was actually quite simple. The documentation on the Forge API is very good, very thorough. It's just a matter of following the steps.

But one place where we had to make some architectural decisions and design this was with how do we structure connecting accounts, and specifically for web applications distributed across people's browsers with a back end server.

And so, just to review the process of connecting your accounts, what the user does is they within Site Scan select Connect to BIM 360. Then in Autodesk they log in and grant permissions to Site Scan to be connected. From that they get an authorization code, which gets saved to our back end server and can be used for subsequent requests.

So, to dive a little bit deeper into how does this authorization process happen, there's these three components of the system. There's the web application-- Site Scan Manager. That's JavaScript running in the browser; there's the Site Scan Cloud running on our servers; and, of course, the Forge API.

So when the user goes through this authorization process, they successfully get authorized, they get an authorization code. That gets sent to our servers, which is exchanged with Autodesk for an access token and a refresh token pair. That access token then can be used for any subsequent requests to make authorized requests to Forge.

Well, one consideration that we had here in terms of building a web application that's different than a server-side application is the web app can't know the access token. Because what if, for example, the user may have spyware on their computer and not know about it. It could get access to that access token and have access to user data. And we wanted to prevent against that.

And so we implemented a proxy where all the requests come from Site Scan Manager and proxy through the back end servers where that access token is added on and then sent along to Forge. And that keeps those requests secure.

Another consideration is refreshing. So that access token is only valid for an hour. So after an hour elapses, for the next request, the back end server knows that it needs to refresh the token. So it sends that refresh token to Autodesk and gets a new access token/refresh token pair.

One consideration here is what happens if you have two requests. So say the front end application loads, there's two different components-- one wants to know information about the project, one wants to know the contents of a folder-- and these two requests get fired out?

Well, the back and says refresh is required; the token has expired. And now it'll send two requests to the Forge API. Well, the Forge API will reply back to each of those with a different access token/refresh token pair. And only one of those two pairs is valid. And so we've got a distributed system, a cluster of servers, running our back end. And so how do we know which one is the valid token? And the answer is we can't.

And so, in order to get around this, what we decided to do was implement a policy where only one refresh token request can happen at a time to the Forge API. So now, when there's two requests from Site Scan Manager, the back end will send that refresh token to the Forge API and put a lock in our database saying this user is currently refreshing their token; no future requests are allowed. Then, once that access token/refresh token pair comes back, then the lock is removed and any subsequent requests can go through. And so that enables us to be able to handle concurrency in this distributed web application.

So I've reviewed a lot of the technical details about what made building Site Scan possible. But now I'd like to take a minute to share some of the team values that we've had that carried us through this transition from consumer robotics to enterprise software as a service.

So I first heard these values articulated by my boss at the time, Dan McKinnon, who heard this from an author named Daniel Pink. And the lesson here is that if you have people in creative roles and you want them to be most successful, you need to make sure that they're pursuing mastery, they understand the purpose of why they're doing their work, and they have autonomy, the freedom to choose how they get that work done.

So in the context of 3DR let's consider mastery. When I started at 3DR three years ago, I had a specialization of robotics. This picture on the left here is a robot that me and my team built before 3DR. And now I'm working on web applications.

And there's some similarities, that's for sure. But I had to learn a new programming language, new frameworks to make this happen. And 3DR was a great environment where I was able to focus on my skills and apply them to the changing product.

Purpose. Of course purpose is useful in any context. Let me just check the time. I want to make sure I'm not over time. Does anyone have the time? 52, OK. So we've got eight minutes.

Purpose is important in any company, of course. But as your product is changing from focusing on building drones to building software that uses drone data and makes it easier to use, it's especially important for the product and engineering teams to stay in sync.

So one thing that we do at 3DR to make sure that this happens is we go on field trips. We engage with our customers to make sure that everyone in the company really understands who the customer is and what their needs are.

So this is a photo from when we were on-site with PARIC actually about a month ago in St. Louis. And as Andy talked about, we really dove into their workflows to make sure that Site Scan could be useful to them. And that's really valuable to me as an engineer to make sure that I know that what I'm building is useful to the end customer.

And lastly, autonomy. Autonomy is the freedom to make your own decisions and get the work done in the way that you think is the best. As the product is changing, it can be tempting for managers to be very prescriptive about what are you doing and how does it get done. But when I think about myself and some of the other people at 3DR who have been most successful, we've really had a lot of autonomy in terms of being able to choose how to get work done.

For example, me as an engineer, I really like to have a distraction-free environment where I can just dive into solving technical problems and really get into the weeds. And so last winter I moved to a cabin in the woods and focused on software development and shoveling snow for the majority of the time.

So mastery, purpose, and autonomy are some of the values that have really allowed 3DR to take drones to new heights. They've allowed us to transition from serving individual customers who want to fly drones for fun to serving enterprises who need to get value out of the data. These values have guided us through the transition from building drones that are very technical and finicky to simple and reliable, from a novelty to a dependency. And these values are what's going to guide us through what's next.

So what's next, now that drones are a solved problem almost? Well, an important thing is ecosystems of data, eliminating the silos of different data accumulating in different places and making sure that people have access when they need it. The BIM 360 integration that we've talked about today is just one example of that. And you'll see a lot more in the coming months.

Also, more automation and analytics. And so this applies to the post-processing of the data, making sure that decision-makers can spend less time looking at the data itself and the photos and more time looking at insights about how the data can drive action on-site.

So we're building this future now in collaboration with our customers. So please join me after this talk or at our booth on the exhibition floor and let's build this future together. Thank you.

[APPLAUSE]

So we have just a few moments for questions. But if you'd like to go, I know the keynote is following this, and so feel free. Yes?

AUDIENCE: So the question-- you said-- I'm an FAA certified drone pilot-- you had said with the automation you don't have to do it. Did you mean you've got the air space classifications and optimization, all that stuff?

NICK SPEAL: So the question is about what sort of authorization is required to fly a drone. And it used to be in the past that to fly a drone for commercial purposes, you needed a commercial pilot's license, as if you were flying an airplane. And that's no longer needed. And so now there's a Part 107 test that you can just take online that the gentleman was referring to. And that's the only certification that's needed now to fly a drone. So, yes, you do need that.

AUDIENCE: You still need that?

NICK SPEAL: Yeah.

AUDIENCE: But you don't do the [INAUDIBLE]?

NICK SPEAL: That's correct. And so we have a training program that you can do through our website to prepare for this test. And then you can do this test online with the FAA.

AUDIENCE: Got it. You gave me the impression that you [INAUDIBLE].

NICK SPEAL: Yeah. So I was blowing through that a little bit. Great. Yes, question?

AUDIENCE: For the [INAUDIBLE] and the volume calculations [INAUDIBLE] have you ever done a test to actually measure how much soil is there as a validation of what the application is doing?

NICK SPEAL: Absolutely. And so the traditional way that stockpile measurement is done is through surveying points. And so you might measure the height at the top of the pile and around the pile and you measure certain points. And then you can use software to interpolate between there. And we're getting very consistent numbers with these traditional methods.

Actually, ours aren't exactly the same. And this is because it's more accurate. Because whereas the most you could practically survey with a point measurement on a pile is maybe 10 or 20 points, this is getting thousands of points. And so it's really capturing all the ripples in the surface that one of these other methods can't compete with. So it's absolutely very comparable. But it's that next level of additional accuracy on top of that. Question?

AUDIENCE: You talked about the georeferencing of PDFs. Have you [INAUDIBLE]?

NICK SPEAL: At this time right now, as a user you just have to click the two points that line up and put it on the map. But it's one of the things we're looking at is if some PDFs can already be georeferenced, and can we extract that data and be able to put it on the map. Yes?

AUDIENCE: Are you using the [INAUDIBLE] photogrammetry to collect the data? Because we work on [INAUDIBLE] create a problem.

NICK SPEAL: Yes

AUDIENCE: Workaround for that?

NICK SPEAL: So the question is are we using photogrammetry to collect the data. And yes. So we're using photogrammetry, a camera on the drone, to collect the data.

This can sometimes be an issue if there's a lot of trees and if the trees are moving, for example, in the wind in-between photos. And so there's some limitations of that. But in general we've found it works quite well.