Behind the scenes: Crafting America's next-gen drone
Delve deep into the process of developing our latest drone, the Skydio X10, guided by user insights and learnings from the team members who steered this project from concept to completion. This panel discussion offers attendees a genuine behind-the-scenes look. Learn directly from the engineers and designers about the hurdles surmounted, the strategic decisions that shaped our path, and the candid learnings gleaned along the way. See the nuts and bolts of drone innovation, stripped of any gloss, directly from those who lived through the experience.
Learn more about Skydio X10.
- To today's panel discussion, um, focused on really providing a
- behind-the-scenes
- view into the development of the next-gen drone, which is Skydio X10,
- which is, uh, designed and built here inside the US.
- Um, I'm Jason Tillman, as, as what was alluded earlier, product marketing
- director over here at Skydio, and very excited to be joined by several
- members
- of our X10 development team. So today we'll be sharing some
- experiences and learnings of our panelists along the way, from the
- conceptualization of the X10 all the way through the release.
- So kind of provide that behind-the-scenes view for you.
- And as you noticed, one of our panelists, Kevin Clemens, was unfortunately unable
- to attend today, uh, but we have plenty of, of, uh,
- conversation with
- our existing panelists, um, so we are not gonna be lacking today
- content or topics today. And so before I hand the microphone to
- each one of our
- panelists for introductions, I want to provide a very quick overview of
- the
- X10 just to set the stage for today's discussion.
- And so September 2023, we announced the X10 at our first
- annual events called Skydio Attends, um, with shipments
- starting later last year around November.
- And a few months later, we released a highly secure version called
- the X10 for
- defense use cases. And there were three areas we primarily wanted to
- focus on with
- the X10, which aligned to the needs that we heard for the market.
- Number one was camera improvements, number two was maximizing the capabilities
- of
- AI, number three was resiliency and adaptability.
- And so first we want to focus on, um, one of the most important factors,
- and that's
- the ability to capture aerial data.
- And so with this we introduce, um, sensor packages, which
- are really collections
- of market leading cameras and senses- sensors for tackling the most common use
- cases, and really allows the X10 to punch well
- above its weight class.
- And these include narrow, wide, telephoto, and radiometric thermal options.
- The next, Skydio's been known for years for having
- some of the smartest drones
- inside the sky, powered by autonomy.
- And we took this to really a new level with the X10,
- and really introduced 10 times
- that compute power on board. And that unlocks all sorts of things, such as,
- uh, the ability to build 3D maps on the drone
- in the field without
- connection to a computer, or autonomously navigate even in complete
- darkness.
- And then finally, drones are expected to be resilient inside some
- of the harshest
- conditions. So we not only made the X10, um, capable of
- thriving in all types of weather, um, with an IP55 rating, but we also made
- it highly adaptable with attachments that can extend the drone's usefulness no
- matter what the mission range requires.
- So we, uh, also doubled down on connectivity,
- which enabled longer distance
- connections. And with 5G LTE connectivity, operators can really
- fly anywhere they want. So, uh, the mission's taking place in Los Angeles
- and the
- operator's over in New York, they can absolutely do that.
- And so, again, allows a level of flexibility there.
- And so with that, it's time to introduce our panelists.
- And I'll hand it off to them for some quick introductions before we move on to
- the
- questions here. So I'll, I'll go ahead.
- And starting with Harrison, do you want to introduce yourself?
- Yeah. Hey, uh, I'm Harrison. Uh, I'm an autonomy engineer here at Skydio.
- Been here for around five years at this point.
- And, uh, I was the autonomy lead for the X10
- program.
- And then, uh, Russell.
- Yeah. Hi, everyone. My name's Russell Bondi.
- Uh, I'm a senior manager on the image quality
- team.
- Um, uh, really excited to be here today, and I, I led the, the effort in, in
- putting out high quality imagery and videos for the, the X10 program.
- And then finally, uh, Asher.
- Hello, everyone. I'm Asher. I'm, uh,
- the mechanical lead for R47, and I started at Skydio after graduating
- school in 2016. So been here a minute.
- All right. Well, excellent. So I'll just turn off the, uh,
- screen share here so we
- can see all the panelists. And so on here, we'll go ahead and get started here
- with the first question. Um, and this is really focused towards Russell.
- Um, so can you share some insights into the
- camera selection process for the
- X10? So specifically, what kind of cameras were chosen and why?
- And then additionally, if there's any valuable learnings you
- gained along the way.
- And most importantly, how do those camera choices align with
- those common use ca-
- cases that the drone actually works for?
- Yeah. Um, it's a, it's a great question.
- You know, and, and to, to paint the, the picture of the X10 camera
- journey I'll- I'm gonna kind of start here at the
- beginning.
- And, and a lot of it was, was driven by, uh, having a
- really great relationship with some of our customers.
- Um, the S2, when we started the X10 journey,
- was out in the wild.
- People were using it. Our customers were, were loving it.
- And, um, and, and, you know, they were really kind of customers and partners
- with us, and they,
- they allowed us to come out and understand the use cases.
- So we had kind of, you know, different key customers in the,
- the different sectors
- and use cases that we were, we were targeting.
- Um, and they were, you know, champions for us.
- And they really were like, "Hey, you know, this is a great product.
- Here are kind of the things we love about it.
- Here are the areas we, we want to improve on." Um, and they, they
- gave us opportunities to come out and visit with them,
- which my team really jumped
- on. We went out to many different customer
- sites and, and watched their, their kind
- of workflow and tried to understand exactly what kind of data they're, they're
- trying to gather and, and what was the, the pain points and how, you know,
- how they did it and, and, you know, what were the challenges.
- And I remember one experience.
- I went on a... I went to a utility company,
- and they actually set me up with a,
- with a hard hat and a, and a set of headphones and a microphone,
- and I was, like,
- plugged into this whole system. And I was able to kind of watch the
- operators go through the workflow of these different shot lists they had and, and
- ask questions over the, the microphone, being like, "What, what exactly are you
- looking at here?" And they'd be like, "Oh, this shot is, is geared towards trying
- to understand, like, the rot on the pole," or,
- "We're looking for the, the neutral
- line in this one." And,
- and so, you know, I think that was a big part of helping us choose
- hardware
- and, and drive our decisions, was understanding what,
- what our customers really
- wanted to do, what data they were, they were truly after.
- Um, and so using that, it was also difficult because,
- uh, customers had a lot of different
- requirements across different use cases.
- And so our approach was, you know, what's the, the hardest requirement
- or what's
- the smallest kind of object that- ...
- that the utility space needs to, to be able to capture.
- Um, so we kind of aimed at that. We'd be like, "Hey, utility space,
- they need to be
- able to, you know, read this tiny text on this plate."
- And we assumed if, if we can
- read the tiny text on that plate, we can probably do all of their, you know,
- requirements or pass all their requirements.
- And same thing for the situational awareness use cases.
- Like, "Hey, what's the hardest thing you guys
- are trying to do?" And we'll tackle
- that one and, and kind of make this assumption
- that we're going to, um, be
- successful with the other requirements.
- And so n- understanding the, you know, what data was needed
- allowed us to then go back and choose sensors.
- We then, you know, worked with Sony and we'd be like, "Hey, you know,
- quickly know
- that, that we need a certain resolution in order
- to accomplish these, this, this
- data capture." And then, you know, once we had our slens-
- sensors narrowed down,
- we moved on to, to lens vendors and we worked with them.
- We'd be like, "Hey, you know, what's off the shelf here?
- What, what can we find?" And, and you know, we also understood
- that this is kind of
- a, a new use case, especially trying to accomplish so much,
- um, with, with the single drone that's in this kind
- of weight class.
- Uh, we wanted to keep everything small and light.
- Um, and so we can't explode our cameras 'cause
- then that balloons the drone that's
- behind it. Uh, so we, we worked with lens vendors
- and it turned out like no one
- had really made these lenses yet. So we, we worked with them.
- We designed custom lenses, uh, really specific for the use cases,
- and we worked with them constantly.
- A huge amount of back and forth to dial in this lens
- and make sure, you know,
- we had, you know, a good amount of sharpness in the center,
- in the corners, and
- that we weren't, you know, contributing a huge amount of flare.
- And there's all these aspects to designing a lens to match the sensor and create
- a really like harmonious camera system that, that works well.
- So we worked with vendors and we, we dialed in this process.
- And then, you know, during, during the whole thing,
- we continued to visit our
- customers and, and really as we, we dialed in our hardware,
- we got comfortable with
- that. Um, we had a, a great sensor, a great lens, an actuator that was able to
- drive it for autofocus. Um, then we stepped up our game into our
- tuning and,
- and that was again, where our customers really helped us
- and we would go out and be
- like with a, with a prototype,
- image their, you know, whatever the dataset they wanted to
- capture, review the
- images with them, be like, "You know, where can we improve here?" And they, you
- know, some examples that come to mind from, from the R-
- or from the X10,
- uh, program are, uh,
- are we went out to a customer and, and there was just a lot of, uh...
- At the current phase, this was like early in the development, we
- were doing, you
- know, denoise, which is like pretty standard in a,
- in a camera system.
- There, there's noise that's being produced from
- the sensor and, and, and you try to
- remove some of that. And we were out with a vendor and they were like,
- "Look, you
- know, you guys are doing denoise, and we, we can see that.
- It's a, it's a pretty n- noticeable, um, effect." And they're like, "You know, we
- don't really care about noise. Like, you know, as long as
- it, it's like there. It's not like a, it's not a consumer product.
- We don't care about noise. You don't have to remove it." Because what
- happens
- sometimes when you remove noise is you actually remove some of the
- information
- that's there as well. Like, maybe there's text on a plate and
- that algorithm
- misidentifies some of the text as noise and, and smooths it out.
- And so from that like learning process, we were like, "Oh, we, we can turn our
- noise reduction way down. Customers aren't really like super focused
- on noise.
- They're more focused on collecting data a- and getting good data." And,
- and so we, we, we dialed things in. We reduced noise reduction.
- We let more noise through, but in the end of the day,
- we had a huge amount of
- information and the data was always there, was we were preserving this data for our
- customers. Um, and, and you know, I think that was super valuable for
- us. I think an end product, it's really valuable for our customers
- 'cause we're,
- they're getting what they want. Like they want information,
- they wanna see cracks,
- they wanna read small text. Um, and so we, we did this
- continued iteration with our customers and it,
- it was really fantastic process for
- us. Um, we ran into, to other things where we wanted to
- add extra layers of algorithms, like our autofocus algorithm,
- um, co- was tricky to, to work with.
- And, and we worked with our customers and found out that like they wanna
- prioritize
- foreground a lot more. So we added a layer there
- that allows us to have multiple
- windows in a center window and really focus on exactly what we want,
- you know?
- It's tricky to, to look at a scene and, and you know, your eye can be like, "I
- wanna focus on that," and then you would normally touch.
- But we wanted to take that touch interface out.
- We didn't want our customers to have to touch the screen every time.
- We wanted our autofocus algorithm to be smart enough to like understand
- the scene, detect exactly what they want
- and focus for them.
- Um, and so those types of challenges were super tricky and, and you know,
- luckily we, we, we caught them early in our process
- and figured out and again,
- that's a lot of like feedback and, and going out and testing and we,
- we really put
- the, the X10 program through its paces. A lot of different cameras.
- We wanted to test them in a lot of different environments.
- Um, all the use cases across the board.
- And so we, w- we've spent a lot of time out in the
- field, a lot of time with, with
- customers and partners, um, dialing in the process, understanding,
- you know, the
- limitations and, and trying to make it, um,
- as good of a ca- data capture platform as we could.
- Um, and I'm, I'm really excited how it turned out.
- I, I think people are happy with it and I'm excited to, to continue working on
- the, the X10 program, you know, moving forward,
- coming out with new releases, new
- algorithms, um, to, to boost the camera performance.
- Um, and I think that, that kind of summarizes the journey.
- You know, we started with understanding what needed
- to be done.
- We moved into this really fantastic hardware phase of lens design, sensor
- selection, you know, choosing the best of the best for,
- for this high quality while
- keeping everything nice and tight and compact.
- And then went into the tuning side where, where I think Skydio really strives.
- We, we listen to the feedback from customers.
- Um, we take a, spend a lot of time doing fine-tuning.
- Um, you know, that, that noise I mentioned, noise to,
- to detail ratio is something
- that takes a, a long process to get it really dialed in.
- Um, and a lot of other aspects as well.
- So I'm, I'm super happy with where we are on the X10 camera systems
- and, uh, it was a really fun project.
- I hope that, uh, that answers your question, Jason.
- Yeah, and absolutely. Um, and could you give us a reminder in terms
- of what the,
- the sensors and, uh, camera, um, look like on the X10 and
- how we organize those in the sensor package?
- Yeah. So we, um-
- We designed everything so there would-- we'd be, you know, a-as high
- quality performance through, uh, the zoom range as possible.
- So we, we paired our, our forty degree, this is horizontal field view,
- um, sixty-four megapixel narrow camera, we call it the narrow, with our tele
- camera. So you can gain the situational awareness,
- the kind of overarching
- view with your forty degree, and then you, you zoom over to your tele which is a
- forty-eight megapixel ten horizontal degree, um, field of view,
- and you can really punch in with that one.
- So we-- that's kind of the workhorse for the
- situational awareness.
- It's also super powerful in, in the utilities industry when,
- when the standoff
- distance is required. You need, you know, let's say ten meters so you can keep a
- safe distance from something that's like highly powered and still punch in and get
- that, you know, the, the ID or the serial number off a part
- that needs to be
- replaced. So we like that combination for, for being able to zoom
- in and also gain situational awareness.
- And then the-- for the three hundred L, we, we paired that with a one-inch,
- um, Sony sensor, uh, with an eighty-degree horizontal field of
- view, and
- we paired that with a narrow camera.
- The same camera that's in the, the Z is also in there.
- That allows you to have a little of that punch in range, but also ki--
- a much wider
- field of view for efficiency with mapping and, and three D scan and other products
- like that. So it's, uh, it's fifty megapixels there with a nice
- big pixel on
- that. Um, a great lens in front of it and, uh, and then all those are paired with
- the radiometric thermal as well.
- Excellent. Well, thanks very much.
- And, uh, you know, you can always have like the greatest, uh,
- camera sensor in the
- world, but if the drone makes it very challenging
- to actually fly and actually get
- the stuff done, it makes the, the process of actually doing these
- missions
- oh so much more challenging. So this next question is for really for
- Harrison. Um, and so Skydio really has a history of
- really producing some of the
- smartest drones inside the sky. What the Skydio autonomy
- is really being core to
- everything that we do. So would you mind, uh, delving into the role
- of artificial intelligence, AI, into setting the X10
- apart? And really how does AI really enhance
- the-- its capabilities or if
- there's any surprise use cases that emerged during testing?
- Would love to hear your perspective on it.
- Yeah. I mean, I think like first and foremost when it comes to like
- integrating autonomy, um, into the drone, Skydio has a
- pretty unique approach. Like it's something that we designed from
- day one. You know, when we're whiteboarding out the--
- when we were whiteboarding
- out the concept for X10, um,
- you know, this wasn't some feature we added at the
- last minute.
- You know, from, from day one when we're like at the concept stage we're
- trying to
- figure out, "Hey, how can we incorporate as much like
- autonomy as possible to the
- system? How can we make this the best like
- autonomous flying robot?" Um,
- and then the other thing for X10 when we were designing it is, um,
- we not only want it to be super awesome at release, but to a certain extent we
- want it to be a platform we can iterate on and develop on.
- So like as advances in algorithms come, we can update the platform and
- make sure it's always at the bleeding edge.
- Um, yeah, in terms of,
- I guess, the, the design process, like the, the biggest thing for us is,
- you know, the design goal from the hardware
- perspective, which is the platform that
- all these algorithms build on top of, um, was kind of having
- as much compute as possible on something that can fly.
- We kind of wanted to make a flying supercomputer.
- Um,
- so in that regards, you know, one of the huge things
- that really enables like
- our capability here is the fact that we shipped with a, a NVIDIA Orin
- processor. Um, the Orin kind of in the form factor
- that we have it
- on is NVIDIA's latest and greatest, you know, AI compute chip.
- Um, that was actually kind of an interesting
- story.
- It wasn't even out when we started the X10 program.
- We kind of made a late decision to upgrade the chip that we were shipping with
- halfway through. Um, and, you know, even though that carried a lot of
- extra E work and a lot of like software work to integrate that, we thought it was
- super important to have, you know, the, the latest and greatest hardware because
- it would really-- it's really like the workhorse
- that enables all the cool
- AI work that we're doing on top. Um, and then in terms of the kind of
- algorithmic side, we kind of think of the autonomy system in
- two different
- layers.
- Um, so the kind of base layer is like our core
- navigation engine, which is what, uh, allows us
- to do like obstacle avoidance, what, you know, the control system
- and kind of the
- visual navigation system that allows us to fly without GPS.
- And I think the benefits of that were the super known
- quantity from S two-- and from our experience with S two
- and X two.
- Like this is kind of the core system that lets the drone fly itself,
- um, and really reduce, you know, the amount of operator,
- uh, like load. And then I think the other thing is that,
- um, the autonomy
- system's also what allows us to fly in really challenging environments like, you
- know, GPS denied environments that other systems just can't fly in.
- Um, and our goal kind of with X10 was, you know, let's take everything with S two
- and X two, um, and that we can already do and make it better.
- So, um, basically in every category that customer-- like in
- our core flight, uh,
- system, we wanted to upgrade it. So this really is like a next generation
- leap in,
- you know, obstacle avoidance and controls and, and navigation.
- And then I think the other huge thing that maybe people don't realize is
- like an AI capability is, you know, the expansion of what we
- wanted,
- uh, X10 to be able to do. So for example, supporting, uh,
- payloads,
- um, I think that's actually maybe one of the
- great examples of like in some other
- systems when you put on a payload, you don't have obstacle avoidance, the
- aircraft's gonna be more challenging to fly.
- You know, you don't get OA, so it's super easy to damage the aircraft.
- So, you know, a huge focus for us for X10 was like
- we knew kind of already the amount of benefits of, of the autonomy system, so
- making sure that integrated well with payload so, you know,
- you can modularly plug
- stuff in, uh, plug in attachments and still have, you know, that obstacle
- avoidance, still have that robust control.
- Um, and yeah, like also kind of recognizing that,
- you know, OA is super useful on S two and X two.
- Unfortunately, that's kind of limited to the daytime.
- You know, we're a
- vision-based system at, at the end of the day, and obviously
- if you have low light
- that can be very challenging. So kind of not just working with the
- hardware team
- to develop the night sense hardware, but also kind of tuning our algorithms to
- work
- really robustly in that environment was a huge focus.
- Um, but that all is kind of the navigation
- engine layer.
- That's kind of, you know, the AI pilot.
- Um, but the, the real thing that's we also focused on
- in the autonomy team is also
- how do we
- have, you know, the, the X10 help you with your mission?
- Um, and, and what kind of applications can we build
- on top of that
- pilot, a stack that we've already built?
- You know, basically, very few people are flying just 'cause they wanna fly for
- the
- sake of flying. You know, people have a mission
- that they're trying to accomplish,
- whether it's to inspect an asset, whether it's to provide situational
- awareness.
- Um, so on the autonomy side, yeah, we definitely for all those
- cases try to,
- you know, leverage AI to make that as easy as possible.
- You know, for situational awareness, for example, developing, you know, s-
- super
- long range, um,
- subject tracking to take advantage of that sweet zoom capability that's provided
- by the sensor package. Um, and on the inspection side, having,
- uh, systems in play like three D scan that allow you
- to automate data capture, automate data twinning.
- Um, and, you know, really ultimately have X10 be, uh, a
- partner, um,
- with folks that are trying to get stuff done.
- Well, excellent. Well, thanks very much, uh, Harrison.
- Uh, yeah, I know that AI is, uh, you know, the buzzword du jour going out there
- right now. And so, um, you know, when...
- I know when we first, uh, came out with our initial Skydio drones it
- was very,
- very new. AI was still very new out there, so it's exciting to see where this has
- actually taken us. Um, but, you know, you can have the best cameras and
- sensors on a drone, you can have it so it's,
- it's amazement autonomy, but if it
- can't really handle what's actually the environment that's being thrown into it or
- can't just really adapt, makes things very, very k-
- kind of challenging.
- So, um, Asher, you know, the physical design of the drone actually
- matter, matters
- immensely here. And so how did the team approach designing
- the
- actual airframe of the X10? So, like, what kind of considerations, um, went
- into ensuring its adaptability, uh, for whatever mission happens to go out
- there?
- And then were there any trade-offs or unique challenges
- that you kind of ran into
- along this process?
- Yeah, good question. So short answer is, like, it was extremely challenging, 'cause
- we designed one drone that's gonna meet the needs of many customers, and many of
- the requirements are the same. Everyone wants a drone to fly longer,
- be m- more
- robust. But, like, the relative importance of each one can
- change a bit for
- each customer.
- Um, so as Russell said, what we do, it's, like, very highly valued at Skydio,
- is working with customers and learning, you know, what do they care
- about. Um, so we encourage customer visits by every-
- not just
- solutions engineers who will work with particular customers with their,
- with their
- trade-offs. You'll... We will go out and, like, sit there and listen.
- And y- you find yourself asking these questions
- that just would never come up in an
- engineering lab. Um, in the last visit I had, I,
- I learned a ton
- about how, like, this one particular customer who
- was using our dual
- charger, throwing it in the back of a car
- and putting a rock on it, and that was
- how they would charge our batteries between two, between two,
- um, sites. Was like, I had no idea this is what...
- this is exactly what was going on. Um,
- so really, it really makes you think.
- Um, so to your question, how do we make the, the drone adaptable?
- Um, when you start designing like this, you get the customer insight.
- But then what you start is just a blank screen with all your
- sensors and
- electronics. This is what Harrison was talking about.
- This is, this is the, the, the early work in figuring out exactly
- what to
- make, and it's totally linked between autonomy
- and hardware
- and sensing of the cameras. Um, and what we do is we move everything
- around in the space and come up with different architectures.
- What's the bal- what's the... what are the benefits of each one?
- Like, uh, should we put the certain... these, these circuit boards over here?
- Should we make them one big circuit board, two little ones? How big are the...
- is the sensor package? As Russell was talking about, like, which
- combinations of sensors would you, would you install?
- Depending on exactly that geometry, how big does the drone end up getting?
- And eventually we start narrowing in on some after a bunch of prototyping
- and, and a bunch of, "Oh, man, this really isn't working." Um, so
- eventually we start narrowing down on s- on, on an approach, and that k-
- kind of comes down to two things. One of them is figuring out which
- capabilities, um, we should turn into attachments.
- Uh, the, the slide you had opened a second ago
- was highlighting all the attachments
- we've had. Those are capabilities that would work really well for one
- customer, but
- we don't wanna, like, s- stick, like, maybe the night autonomy package for
- every customer. That's heavy, and certain customers
- are always just gonna be
- operating during the day.
- Um, so our design approach is, one,
- pull off all these unique sensor... th- these unique capabilities.
- And the second one is what's on the core drone,
- it's just this balance, this
- continuous trade-off of sensing, flight time, how
- big it is, the size, um, and the general robustness.
- Um, and we wanna just nail those core
- capabilities and the, the balance between
- them. So the sensing, as, as, as Russell was talking about,
- is the size of the u- user cameras, which combinations.
- But also, how big is the w- the GPS antenna?
- How accurate and reliable do we need the GPS to be?
- Everyone wants that to be really good, but, you know, you're not gonna make your
- GPS huge. Um, same for WiFi antennas.
- Range is, is critically important, but it, it...
- we can't just put, like, your, your home, home WiFi antennas onto the drone.
- So there's a s- there's a, there's a size balance there.
- And then our core chips, we went top of the line NVIDIA processor
- with the Orin,
- but, but we also need to set thermal limits.
- So I can make a heat sink that's super big and, uh, now I can, I can run this,
- you know, on, on the s- probably not in the sun, couldn't get
- that good, but I
- could run it really hot, and I want to trade off the thermal limits
- with, you know,
- the size of the drone.
- So you're balancing the sensors we put in there with the size.
- Certain customers, they're gonna pull the drone out of their
- trunk, they don't
- really care about the size. Others wanna have it in a backpack folded
- up, so
- every bit, every tr- little trade-off we make will slowly grow
- the drone. Uh, like, a good example of that was, was...
- I keep mentioning the, the, the camera packages.
- We were working over and over on this with, with Russell.
- We f- we went through-
- Like a dozen different variations of which cameras we want.
- And I think at the end we've really nailed it.
- But each one you're like, "Okay, we want two cameras next to each other."
- Like the
- gap between the cameras
- will define the size of the drone overall.
- And if you add more cameras, then the gimbal gets heavier,
- you kind of need more
- sway clearance to have good footage, and then that keeps making the drone,
- um,
- bigger and bigger.
- Um, so I've mentioned, like, sensing the size and packing everything in
- really dense. Um, and the last big variable is
- flight time. It just comes down to every step of the
- way, where can we
- take out mass?
- Uh, and it's ... Sometimes it gets ridiculous,
- but each bit we know, like,
- this is gonna be a customer using this drone for thousands of hours.
- You know, every, every additional headroom on flight time,
- uh, you know, that
- matters.
- Um, and the, the last thing is always robustness.
- So the drone
- is a, is an enterprise drone. This is like built ground-up for these
- enterprise use
- cases, which means it's...
- We need it to work whether it's in, you know, the back of a, of a
- car for public safety, if someone's driving it around all day for
- inspecting energy utilities or if it's in the de- defense base.
- You need it to work every time and just 'cause you,
- you know, drop it a little bit or, or handle it roughly,
- that's what your job is
- and it needs to work. So the whole, our whole entire
- design process is balancing, balancing these variables
- and moving around until we
- finally converged on what the n- X10 looks like now, which I
- think it's gorgeous.
- Absolutely. Well, thanks for that perspective there, Asher.
- Um, and you know, uh, it gets brought up, you know, we, we talked about some of the
- prior drones that we've actually had here. So this is a question for Harrison.
- Is it a pro- You know, when we look at some of the functionality
- that was in prior
- drones, like let's say the X2, for example. Let's say autonomy in the X2.
- It could be pretty much anything else in there.
- Is this simply a process of actually just moving that over to the new drone, like
- the X10? Is it that simple or is it a little bit more complicated than
- that?
- Yeah, that's a good question. Um, yeah, I feel like to a certain extent we get
- asked this internally a lot like, "Hey, you made S2 and X2 fly great.
- Shouldn't you just have to port some code and make it work?" Um, I think, I think
- the real answer is, is that, yeah, it's actually surprisingly a lot of
- work. Um, or not so surprising when you think about it.
- Like our, our autonomy stack is really tightly coupled, um,
- and really kind of designed to max out the performance of our hardware
- platform. Um, and X10 really does kind of represent a
- huge kind of hardware architecture shift from S2 and X2.
- So I think, you know, there was a lot of areas going into the program
- where
- quite frankly, like we were kind of nervous whether
- or not we'd be able to even
- meet S2 and X2 performance, um, because kind of at a high level S2 and
- X2, um, made a lot of design trade-offs to make
- the autonomy problem
- easier. Um, and for X10 we really wanted to not just
- have best in class
- autonomy, but like a best in class aircraft.
- Um, like as a specific example with our nav
- cam geometry,
- uh, those of you that are familiar with S2 and X2 will note
- that like the nav
- cams are spaced very far apart. They're on the, uh, basically,
- um, out on the end of the arms by the
- propellers.
- Um, from a robotics perspective, this is great 'cause this creates like what's
- called like a lot baseline and algorithmically you want as much
- baseline as
- possible to have good, uh, OA performance.
- For X10, um, but you know, moving the cameras out there has a lot
- of mechanical design trade-offs. You know, you have to make your arm super stiff.
- There's a huge mass and flight time penalty associated with
- that.
- It's a lot less robust, um, to have the cameras kind of just
- out there. It makes them a lot more susceptible to,
- to damage.
- So for X10 we wanted to move it, um,
- back on to the, the vehicle body, but that presented a
- huge host of kind of autonomy problems in the sense that, hey,
- like your autonomy algorithm has to be
- basically four or five times better just to match the performance of
- S2 and X2, um, because of that reduced baseline.
- And oh hey, by the way, on top of that we're not just trying to match
- performance,
- we're trying to make it a lot better.
- So, um,
- yeah, I think that's an example of, yeah, we had to spend a lot of effort and, you
- know, thankfully we were able to figure that out, um,
- you know, make a lot of
- algorithmic and processing improvements and ultimately get better performance.
- But I think that kind of goes to show like how kind of tightly coupled, um,
- some things that seem like, oh, it should be easy to port actually
- are to like new
- hardware.
- Um, and then, and then Russell from a camera
- perspective, have there been any more
- advanced functionality, um, that we've enabled that really unlocks
- additional depth inside some of the use cases?
- Um, yeah, I mean, I think just kind of, you know, piggybacking
- on, on what Harrison was just talking about,
- like designing a next gen
- navigation camera that again, like he talked about the reduced baseline.
- You know, when you reduce the baseline there's...
- You have to improve your camera systems.
- Um, and so we had to take that and, and really, you know, the
- unique piece of the nav camera is that a standard camera you always want the
- sharpest part to be in the center of the image, and that's kind of how everything
- was designed, um, going, you know, for, for all our generations before.
- And so when we looked at the, the X10 navigation camera,
- we had this reduced
- baseline. We needed to add a new layer to our
- lenses.
- So we designed a really specific navigation lens to have super high
- performance quality at the very periphery of the lens because our
- lenses are, are, are kind of positioned like this on the
- drone, right?
- Like the, the center of the lens is looking at the sky and the ground,
- which really
- isn't the key, right? We wanna be looking forward towards trees
- and branches and
- poles and whatnot. So you're always looking at the,
- the real extreme part of the
- lens, which especially on a really wide angle
- lens can be degraded.
- So we designed our lens specifically to have a super high,
- um, pixel density along that edge there to give us better, um,
- image quality,
- more information, more data to help kind of mitigate
- that reduced baseline.
- And that's just for the, for the nav camera.
- That was a, a cool functionality that we added in.
- Um, and then for the user cameras, the sensors we selected were super
- versatile sensors, right? Like we had a, a, a ZZ HDR
- on the S2 and X2 program that, that was great, but it came with a
- lot of downside. Um, and then we moved over to a quad Bayer
- HDR,
- which is a, a, a different type of sensor.
- It allows you to have- Um, multiple exposures per single
- pixel. Um, and so you can get this really high
- quality,
- wide dynamic range with, uh, a really great image quality still.
- You're not taking a huge amount of loss there because the ZZ HDR, there was some,
- some loss in that. Um, whereas the Claude Bayer is a,
- is just a better algorithm
- and, and a really cool way to stretch the
- dynamic range a huge amount without
- losing much. Um, we also moved up to the sensors that
- are a much higher
- megapixels, right? Like 64, 50 and 48.
- Like, that's a lot of megapixels to be cramming
- into such a small package.
- And then pairing that down with, with a binned mode
- and a low light mode that, I
- think we call it quarter low light, is super powerful.
- So if, if you're in broad daylight, this camera, you fire it up,
- you can take these
- super high resolution images. And then let's say
- you're doing a bridge inspection or you're flying, you know, post sunset,
- we have
- this low light mode, which allows us to, to make the pixels bigger, stretch the
- exposure times and do some functionality there.
- So I think the, the coolest kind of advancements there
- are just the, the
- functionality of these sensors, the versatility of them, um, is,
- is... it was a really powerful and exciting part.
- And again, we- we're pairing that with, with a bunch of, um, fancy components like
- lenses and sensors makes this, you know, pretty powerful, uh, camera
- packages.
- And, you know, uh, Russell, just I would be amiss
- if I didn't mention that the, uh,
- the recent, uh, testing that we've done with Imatest to actually
- showcase those.
- 'Cause, you know, that's, that's been one of the ones that,
- you know, we've
- actually heard, quite frankly, in the past is, you know, your guys', um,
- cameras,
- they're, they're good, but we really want more.
- We would demand more from that.
- And so, um, I don't know if you wanna go a little bit into
- that Imatest, uh,
- testing, um, and just kind of provide a little
- perspective on that.
- Yeah, absolutely. I mean, uh, you know, we worked with customers a lot to get
- their
- feedback, but another way we kind of approached
- solving and making sure that our
- data was, was high enough quality was, was understanding what they
- were using today
- that was getting the job done. So we'd ask them like, "Hey, what,
- what vehicle is
- flying out there today that's getting you the data that you need?" We'd then go out
- and buy that vehicle. We'd then benchmark against
- that vehicle being like, okay,
- the X10 cameras are... You know, our kind of goal was to s-
- to find the industry
- standard in all the places, and then make sure that we exceeded it.
- So we'd go out, we'd buy the, the, the drone.
- So that was kind of our in- internal process, right?
- Using these charts in the background, and you can see behind me, um,
- we measured,
- like, the competition. We made sure that we were sharper and better,
- and then we
- turned it around and went outside and subjectively.
- And then we, you know, we partnered with Imatest, who,
- who's a fantastic, um,
- you know, camera company, camera, like, assessment company.
- They, they build a lot of charts and, and, and software to, to analyze and
- understand, like, the camera performance.
- Um, and we sent them, you know, the, the, the product that we had heard from
- customers were, like, best in class, and we also sent our, our X10.
- And they were able to measure, you know, the key aspects of a camera system.
- They measured the sharpness, which I talked about.
- And sharpness comes down to, to noise performance
- and how good your system is,
- 'cause they were, they were measuring JPEGs.
- Um, the color accuracy, dynamic range, and then noise performance.
- And, you know, we, we thought that that data that came back from them was
- fantastic. It showed X10 really kind of, uh, you know, a good, a
- good amount ahead of the competition.
- Um, and then, you know, the part that to me that made the most sense
- is that we
- weren't, like, defying physics. Like, there was areas in that report
- that made it
- very valid. Like, if you have a much larger pixel in a
- really
- low light scene, it's gonna accept more light.
- You should see the data kinda swip- swap, you know, and things should, should come
- short. So, like, full res should do worse, and it all kind of matched our
- expectations. Um, so it l- allowed Skydio to show that we're
- really a high-class camera that's like, you know, exceeding, uh, the
- competition. But also it wasn't just like, you know, making Skydio seem like
- this absolutely ridiculous system.
- Like, I think it was very level and, and a very honest report, which I,
- you know,
- uh, encourage people to go out and, you know, maybe not read the, you know,
- the
- whole report 'cause it's a little bit dense.
- But at least check out the, the blog post and, and read through that.
- I think it's a pretty good summary there, so...
- Absolutely. Um, and then, um, th- this actually relates to one of the
- questions we
- actually had submitted here. Um, and this is a question for you, Asher, is,
- you
- know, many of you may be aware that we design and manufacture the drones
- here in
- the Bay Area. So specifically, we manufacture in our facility in Hayward,
- uh,
- California. So Asher, um, what are the advantages and
- challenges of doing this?
- So just like the, the, like, design part that's really linked between
- the software, the sensing, and the me- the mechanical team, we, we
- discovered pretty early on it was gonna be the same for manufacturing.
- That, you know, for, we have this
- special navigation camera designed to work really well on our drone.
- We're gonna wanna own that every step of the way.
- So how do we res- like, we can get now the engineers on site
- when we're
- receiving cameras, inspecting them.
- What does the fixturing look like when you assemble it into a product?
- And what does the
- testing look like afterwards? And then how do we package it up
- and put it, put it
- into the product? And we, we had all these really interesting
- learnings early on
- where we discovered, oh, we, you know, we really shouldn't be doing it that way
- 'cause that's not protecting the sensor of the camera.
- Or you, or you have an image... Or you have the,
- uh, camera design engineer sitting there on
- the line s- like helping design
- exactly how w- we'll test it and make sure that we not only bought a
- good module,
- but we put this good camera module into the drone in the right way,
- and it's gonna work the way we expect.
- And that applies on the camera level.
- So it also just applies on the mechanical design level, most of wh- what
- we work on. So I get to own, not... Well, I get to be involved in
- both designing product, the fixtures that are used to
- assemble it, and then working with the exact operators
- who are on the
- line as- who are on the line putting it together,
- you know, soldering the wires
- down. Um, and with that control all the way across the, ac-
- across the
- process, we can make sure we are not only designing but
- manufacturing and taping up and shipping, you know, the best product possible
- that's gonna meet the needs, like the high, the high-quality needs
- that our
- industry customers require.
- Uh, and then, uh, Asher, you know, you mentioned, uh,
- earlier about reliability
- needed for a drone of this caliber.
- Can you, uh, share a little bit about the testing
- that went through to just make
- sure that it was ready for prime time and getting into the hands of customers?
- Uh, yeah, sure. Like, um, we try to put the drone
- through
- everything you can imagine that the, the corner case customer's gonna do.
- So like you're going, it's late night, you're working, and you open the fridge,
- and there's like a drone in there getting cold before a test because all the other
- ovens are, ovens are being used. Just across the board we
- just try to do everything
- to push the drone to its limits. We make it hot, we make it cold.
- We put it in high humidity, low humidity.
- It's been off, it's been in the freezer.
- You take it out, you try to fly it right away. Hot sun.
- Up in the mountains.
- You know, we do thousands of hours of these
- testings.
- Um, there's some cool videos that our reliability team likes to make
- with everything cycling and moving in a room.
- Kinda
- looks like a li- attack of the robots, but just
- in a room and cycling and trying to find everything that fails.
- And that's what we do. Something fails, we're like, "All right,
- this one interface,
- this one bracket isn't quite right.
- We can change this on the next revision." That's how hardwa- hardware development
- works. You push, you push the product to its limits,
- you find the failures, you fix
- the failures, and you, you test again.
- Ex- excellent. Um, well, uh, let's move on to some of the questions we,
- we've had,
- uh, submitted so far. So the first one is, uh,
- you know, a nice little warm-up question here is,
- what was the most fun testing
- scenarios for the X10 that you've done?
- Uh, for me, I was like pretty involved in a lot of the
- flight testing, like with
- the autonomy bring up and the controls. So I think we have a...
- You know, at the end of the day, we, you know, if we say we're gonna,
- for example,
- qualify th- the, the product to fly at like 15,000 feet
- density altitude, we
- will actually go to 15,000 feet density altitude and fly a drone.
- And for me, that personally, that was like super fun.
- Um, I'm a huge mountaineer. Uh, that's a huge hobby of mine, so it was
- an interesting opportunity to, you know, combine something I like doing in my free
- time, like hiking and backpacking in the mountains with,
- you know, actual robotics
- and, uh, engineering work.
- I wasn't sure if Asher or Russell, anything from your perspective you thought
- was,
- uh, kind of an interesting testing scenario?
- Yeah, I mean, so one kind of common use case where,
- where cameras can, can
- struggle, um, is in, in, in snow, right? Like it...
- The, uh, white balance algorithm is trying to understand what the gray
- is in
- the scene. That's how it makes its decisions.
- Um, and so a, a really typical challenging scene for
- that i- is, is a snow scene
- where everything's white and then the algorithm's kind of trying to
- look for some
- blue and some green. So, um, we recently in,
- I don't know where everyone's located,
- but here in California we got a huge amount of snow in the, in the Sierras, um,
- and
- I took that as a fantastic opportunity to go up to Tahoe, um, and
- try out the, the drone in, in those conditions and also, you know,
- spend a little,
- a little weekend time getting some, some skiing in.
- So I kind of paired that with like a, a really fun adventure.
- Go get a, you know, some powder runs in, and then also, uh, make sure that our
- camera was robust in those challenging conditions.
- So, uh, yeah, that's just one example.
- All right. Well, perfect. Um, and the next one was, uh,
- can you share a little bit
- about how user feedback was included in the development of the
- X10? So were there any kind of community
- insights that shaped the
- project's direction or func- uh, functionality or features?
- Yeah. I guess one thing for, and it kind of goes to some of the
- questions that are
- coming into the Q&A, like we, we do at every stage, you know,
- reach out with... to customers, like get flight test feedback.
- Um, you know, so we
- bring features that people test in real life.
- I think one of the interesting things of testing in real life was finding out how
- popular like LTE, um, as a feature would be.
- I think going into this program a lot of us were maybe kind of skeptical of it, but
- you know, working with various, um, customers, like
- seeing their use cases and also hearing their pain
- points of, you know, for our first couple prototypes,
- like what are the issues that
- you're seeing? A lot of it was like latency, connectivity,
- reliability related.
- Um, and getting that feedback was super useful.
- But also I guess for us like super informative, 'cause we weren't necessarily
- expecting that to be, uh, as key of a feature as it would've been,
- uh, at the
- beginning of the program.
- And Russell and Asher, anything from your side that you saw,
- any sort of user
- or beta feedback that you thought was actually really helpful in terms of
- development?
- Sure. It's, uh,
- it's sp- uh, the, the first thing that came to mind is our dual charger on
- the new
- product. We wanted to ask customers, I, I mentioned this when I talked about
- that customer visit, like how, how often do you need to charge the
- batteries
- and then how many do you need? Um, and that comes into uptime.
- How long do customers want to be able to operate one drone continuously, and what,
- what does that infrastructure look like?
- Do you need like,
- what, 16 batteries to operate during the day
- or like few batteries and chargers
- that can charge them really quickly? And what does your power supply look like?
- Um, so specifically I visited a local customer
- and saw what that setup looked like
- in that truck to see like, okay, this is, you know, we wanna have a high power
- adapter. So if you look at our dual charger right
- now, there are two
- plugs in the back of it. One of them can plug in a normal 100 watt
- USB-C
- cable, but we also know some users wanna be able
- to charge a battery in under
- an hour.
- Um, or if... Actually, I think it might be close to 45 minutes.
- We also have a 230 watt barrel jack that you can plug in for fast charging,
- and that was a specific capability we added in because we knew customers
- didn't wanna carry a ton of batteries.
- They had the capacity to charge them, they just needed it to be done quickly.
- So that su- defined this new, this new connector and also the size,
- 'cause we
- wanted to add and dump a little bit of heat to pull, pull that much power through
- it.
- Perfect. Well, um, we got a couple of other questions inside
- here.
- And, um, Harrison, this one's for you.
- Do you have an API to allow end users to build use case specific
- add-ons on top of the navigation stack?
- Yeah, that's a good question. We get that a lot.
- There's obviously [laughs] a lot of interest, um, to, to do that.
- Uh, the answer is, so right now the answer is no, we don't.
- Um, and I can actually provide a lot of
- context as for why I think-
- Internally, as we've been developing software for the
- drone, there's kind
- of two aspects to this. One is that, you know,
- our autonomy stack really is this, like, tightly integrated thing.
- And a lot of the times we're, you know, pretty close to
- running to, like, the hardware limits of the system.
- So there isn't--
- Like, at least for S2 and X2 there weren't, uh,
- there wasn't as much
- space to run extra compute. With, with the Orin, obviously
- that opens up a new
- world of possibilities, but that's just kind of explaining why we
- haven't done it
- before. The second is that we've found it's kind of
- hard to make a clean API. Uh, I
- think,
- you know, when you're thinking about a problem,
- it seems, like, pretty
- straightforward. Like, "Oh, I'm gonna have an API.
- Like, maybe I can just, like, define these points to go to."
- But then I think our
- experience actually, like, implementing stuff like 3D scan
- and implementing stuff
- like Housecan, which is a partnership we have with Eagleview,
- is that
- a lot of the times halfway through the process, um, you know, maybe one of the
- APIs isn't as clean as you thought.
- Maybe you need capability that you probably don't want to expose,
- um, because it kind of goes into a lot of
- internal machinery, and it's, like,
- pretty delicate, um, but you need to get your, your task
- done. So, um, from that aspect, I, I don't
- know that,
- you know, if we released an API, if it would actually do what a lot of
- people
- want. Um, and I think the third aspect is, like, just from a pure,
- like, engineering perspective, um,
- w- it, it takes a lot of effort to kind of like
- sandbox the code.
- At the end of the day, we want the system to be really reliable,
- so, you know, w-
- whatever we give to the user, we want it
- to be robust and fly. And then I guess, you know, if you're allowing folks to run
- kind of arbitrary
- code on your platform, it's kind of hard to,
- to protect against that.
- Um, but with all that being said, yeah, I think the...
- that's one thing that we're looking at kind of at a very
- high level with
- X10 because of the extra compute, seeing how we can maybe make compute
- sandboxes and maybe bring that out.
- Um, one thing we are definitely making more
- concrete steps to, though, is, like,
- allowing for API access to the
- data capture. Um, I think we found out
- that at the end of the
- day, most folks that want API access don't
- really care about flying, um, they care about the data,
- right? So we are making a lot of, um, efforts
- to, like, with the, with the cloud product we offer to kind of
- allow API
- access at the imagery and at the data level, but not
- necessarily at the, like, flying robot level.
- It looks like there's another question here, uh, Harrison.
- It's, uh, is there an open interface for the Orin
- for customers to use their own AI
- on the camera and drone data received onboard?
- Yeah. At the moment, uh, no. Like, uh, we don't
- allow that. But again, that's, like, something we're potentially exploring for
- the
- future.
- Um, and then another question came in.
- It says: How secure are the C2 and sensor links versus JDI?
- Are cellular communications providing reliable and good choice for UAS,
- especially in BVLOS expands?
- I'm not sure if we have the expertise on the call here today to address that, so...
- Yeah. I guess I can, I can comment to it to a certain extent.
- I think for the latter question, like, is, is, um,
- cellular providing useful for BVLOS?
- I think one thing to really
- remember is that, you know, a lot of the times a C2 link for, like, a,
- a drone is
- really important because you need control and, um, I guess
- not quite C2, but it's like RF related.
- You know, you need GPS signal to robustly fly.
- I think a huge advantage of the Skydio system is, like, you know, we are an
- autonomous system from the ground up, so we are a lot more robust
- to the,
- you know, to C2 dropouts, to, like, high latency in our C2 link.
- Um, so for us, with our platform, we've found that, yeah,
- cellular is proving to be a great choice.
- It's super useful to get that connectivity option.
- And I think it's something that would almost be impossible to pull
- off actually on
- other vehicle platforms because of all the, like, innate nature
- challenges with cellular, um,
- do kind of make it a crummy C2 link if you don't have an autonomous platform.
- But because we have that autonomy, it really enables this, like,
- secondary, uh, connectivity. Um, in terms of
- security, yeah, like most of our wireless is, like,
- pretty...
- You know, we follow industry best practices.
- I can't comment specifically about the wireless security 'cause that's not
- really my area of expertise. But, like, in terms of overall kind of software
- security, like X10, we really did take it super seriously.
- Like, through our development process, we have a lot of pen testing and,
- you know,
- it's our first product that's leveraging a lot of, like, hardware security
- features that are baked into the, the chips we have to kind of ensure, um,
- integrity of our software load, like from factory to customer.
- So, and, and again, I'm not sure if someone on this, uh,
- panel would be able to
- answer this. This is more about, um, the communications.
- And so, um, Steven actually asked: Can you talk more about the X10 as it
- relates to
- comms, e.g., preventing jamming and the like?
- Um, and that may be a question we just may
- take off-offline.
- Um, but, uh, anybody on the panel have any perspective
- on that?
- Yeah. I mean, I guess kind of related to the previous
- point.
- That's actually kind of a huge...
- That's kind of a huge area we think about in terms of autonomy is how
- can you operate in an RF denied environment?
- Because it doesn't just happen for, you know, defense customers.
- Like, there's plenty of inspection use cases
- where folks wanna go in a factory or
- go in a power station where, um, you have EMI
- and other, other interference that degrades, you know, GPS,
- degrades comms. Um, yeah, I think overall our, our
- kind of strategy
- for X10 is really, again, relying on the visual navigation
- system. I think that unlike basically for other
- drones, right, if you jam the entire RF spectrum, you're,
- you're a little hosed
- because you're relying on GPS and you're relying on a command link to
- stay in the
- air. But, you know, for us- Um, you know, in testing we've had cases where
- the entire radio subsystem fails, but the, the vehicle kind of flies itself and
- returns to home because, you know, at the end of the day, we do have
- that visual
- navigation system. Um, and we do have
- that, like, capability to not have to rely on RF to
- stay in the air.
- Um, obviously we
- put a lot of effort into making those RF links as great as possible, but I think we
- do kind of, because of the autonomy system,
- have a lot more, like,
- intrinsic robustness to those scenarios.
- So even if you do lose a C2, like, like, hey, you can still do your data capture
- and still get back.
- You know, another thing to bring in mind too is
- that there's actually two different
- variations of the X10. There's the X10, which is the one that was, uh,
- announced over in September. Then there's the other one
- that actually just
- recently, uh, came out, and that's the X10D.
- And the D stands for defense, so defense a- applications.
- And that's the offline version. More secure version.
- It uses MicroHard, um, in the controller, the, as
- well as inside the drone itself. And so one of the things we-
- we're introducing
- with that is actually dynamic channel switching.
- And so therefore if- if you are getting jammed on one particular
- channel, it will
- automatically progress to the next channel to make sure you actually get that, um,
- that level of, uh, connection that is appropriate for what you actually need
- at the
- time. So we are looking at even enhancing this
- even further in the future.
- But again, the two different versions, the X10, which is more the offline ver- or
- the online version. The X10D is the offline for defense applications.
- Um, and then another question we had here is,
- um, it's
- related to transmission towers. And so I'm gonna create a 3D model of a
- transmission tower, choosing parameters such as the floor
- and ceiling.
- Will the drone recognize and avoid the cables of the transmission
- line?
- If not, is there a way to do, do it? Um, I can actually talk to this one, um,
- really quick, and I don't know if anybody else has anything on here.
- So, um, uh, we actually, uh, shared something earlier,
- uh, last year about something called, uh, semantic understanding or semantic, um,
- uh, understand the environment. So I think that when you look at things
- such as
- transmission towers and the cables that come from that,
- that's actually really
- important because the drone needs to understand what is actually there.
- It sees the tower, but it also has to understand
- that there's actually cables
- coming from it. And to avoid those, because again,
- part of obstacle avoidance is
- not hitting those high-power, uh, cables.
- And so that's actually one that our team is actually working on.
- And actually, again, we gave, I believe we gave a early-stage preview,
- um, last year on this one. So again, we're actively working on that one.
- But again, understanding what's not only in,
- in front of, also understanding
- what is around it, um, is actually gonna be very important.
- So you should see some more about that later in this year, but again,
- an area we're
- actively, actively working on right now.
- Um, I don't know if anybody from the panel had
- anything else to add to that one,
- but...
- Um, yeah, I guess like when I think about that question,
- I kind of in my head split
- it in two ways, right? Like, the first core question is, hey,
- if I fly around a
- bunch of...
- around a bunch of challenging structures and about, about a bunch of challenging
- environments, will I hit obstacles? And the answer is, uh, no, right?
- Like, the core of the navigation engine is,
- is even, you know, if you're manually
- flying the vehicle around, you'll have that autonomy,
- you'll have that obstacle
- avoidance. And a huge host of our customers, um,
- are already doing
- that, you know, inspecting very similar assets,
- if not transmission towers.
- And, you know, the autonomy and obstacle avoidance
- is a huge reason of why they
- trust X10 to do that kind of inspection. Um, yeah.
- And like you said, the second half of that is like, hey, can I tell it
- scan a transmission tower and have, you know, Skydio automatically figure out,
- um, how to do that? And yeah, like you said, with semantic AI,
- that's kind of what
- we're progressing towards and one of the things
- that we're hoping to show in the
- future with... especially with the compute capability
- that's now available on X10
- because of the Orin system. Um, we can kind of build those higher-level
- systems
- ultimately.
- Um, and another question, um, was, uh, regards to, uh, crosshair
- coordinates. Um, so, "I had the opportunity to see a demo of
- the X10 but I didn't
- see crosshair in manual mode. Is that something that'll be added in a
- future
- software update, or is that already in the drone operation I just
- didn't see it?"
- Um, and yeah, not sure if this is the right audience for
- that one on the panel,
- but, uh, if not, what we can do is that, uh, uh, we'll actually
- follow up with you on this one. So anything we are not able to actually
- answer
- today, we'll make sure we follow up with.
- But I guess that opens up for any other questions that there might be out there.
- And if there are, feel free to submit those in the Q&A.
- Um, it looks like we have one other question
- inside here, and that is, "You're
- competing against a much larger company as, and organization set in the drone
- space. How do you make this product with such a
- small team?"
- Hmm. I've, I've got a comment on that.
- Just
- it's be- by being, like, really focused in terms of what
- capabilities are gonna add value to the customer.
- We get the drone launched sooner,
- make the manufacturing, you know, spin up faster.
- And by focusing just on those, you can, you can do quite well with just
- a small focused team. Just like with a smaller team,
- there's less communication
- overhead. So what we will do is empower and
- empower individuals,
- um, to, to know when they can get, get totally lost in a program that we've
- given them ownership of,
- and, and, and when you need to look out for more
- help.
- Um, so yeah, it comes down to empowering individual,
- individual, uh, engineers.
- Perfect.
- To Russell?
- Um, yeah, I mean, it's... I, I think it's...
- I was gonna,
- you know, go a similar path. I think, I think one thing that's unique about
- Skydio,
- like, if, you know, I've, I've seen people come here
- and interview and, and, you
- know, go through the process and, and something that's really obvious to
- most
- people is that we're-- everyone is, like, really excited
- about the product and really excited about what you're
- working on.
- I think if you have, like, you know, 10 people trying to do something and
- they're... no one's really into it, it takes a long time
- and you can get it done.
- But here you can have like one really fired up engineer working on something,
- and
- they're gonna, like, really put a bunch of effort into it and,
- you know, maybe work
- a little bit longer hours and, like, they're passionate about it.
- It's like, it's almost like, you know, a hobby is at the same time.
- Like, you're just so into what you're working
- on.
- I think that's what I noticed about it is we're, you know, we're really excited
- about each component that we're building.
- You know, if you're, if you're, um, Asher and you're building the hardware,
- Asher
- loves it. Like he, you know, eats and breathes hardware design.
- Um, and me, I love cameras. And so, like, I think we're all passionate about what
- we do. So we have this small team that's all
- fired up, and we're able to like, you
- know, do what 10 people could do with just one
- person just because, you know,
- we have that kind of intensity and, and, um, excitement.
- All right. Well, excellent. Um, I'm fired up just by hearing about that,
- guys.
- Um, and so last question we have here is kind of more of a general one.
- It's that, what were the challenges with consumer
- drones, and why did you choose to
- discontinue?
- Oh, I, I can, I can start. There's, uh, there's a difference between selling,
- like,
- a few things to, like, you know, 10,000 things to 10,000 customers or
- sm- fewer things to, to fewer custom- or the,
- or the same number of drones maybe to
- fewer customers. Um, and the, the value is that, is there is that you can really
- focus on the capabilities you're trying to hit.
- Um, consumer drone space is super fun, super exciting, and I- my understanding
- is eventually we'll want to get back there.
- At least that's the, uh, that's, that's the line I hear,
- um, and that we're working towards.
- Um, but by, by switching to the enterprise space,
- there's, there's so much
- value we can add by focusing here on just changing the way the
- world works.
- Um, and that's something we, we, we wanted to do, we're excited to do.
- We heard from customers that, you know, "If you build it, we, we will use it."
- So that's... Those are my thoughts.
- Yeah, it's actually kind of interesting.
- I, I actually don't know that we'd have as capable of an autonomy
- platform
- if we hadn't focused on consumer first.
- Like, to a certain extent,
- um, you know, if you were to... You know, if, if your goal was
- to provide ISR to asset inspection, and
- then somebody said, "And the way I'm gonna do that
- is by making a robot that can,
- like, totally fly itself and... first, and then, like,
- figure out how to solve your
- problem," um, people would kind of call you crazy,
- but that's kind of, like, the
- approach that we had to take be- from going to consumer first.
- Um, so in, in some ways it's, like, more challenging on the navigation
- engine. Um,
- and
- th- so I think it was actually, like, from, from the autonomy perspective,
- we were
- very excited to kind of pursue that path.
- But I think to
- why we ultimately, like, shifted away, like, you know, I think at the end of the
- day
- we were super excited to see that, hey, by making this,
- like, self-flying platform, it actually makes it perfectly suited for
- all these
- other use cases. Um,
- and
- we kind of wanted to focus our energies there, 'cause I think also ultimately we
- wanted to focus on what had the most impact and, like, yeah,
- there's a bunch of super important use cases out there that, uh, can use this
- technology. And we ultimately wanted to focus our
- efforts that way.
Watch Time
57 Mins 34 Secs
Related Webinars
Beyond drone camera specs: Looking past the gimbal and the quality factors to look for
Watch webinar
Drone as First Responder: Changes in Policing, Changes in Prosecution
Watch webinar
Demo: 1 Remote Operator. 2 Inspections. 1 Response. No Field Dispatch
Watch webinar