Consumer drones: it’s all about integration

I’ve written a lot about how computer vision and sensor fusion in drones are going to be big themes in 2016. DJI’s new Phantom 4 is a great example of a giant canon blast across the bow of the greater drone industry. It’s the first mass-market consumer drone that effectively uses computer vision for both obstacle avoidance and subject following. It’s also the first popular consumer drone to include a redundant IMU and magnetometer (compass). These new features give drones a lot more knowledge about the world and make them more safe and reliable, which is very cool, but the thing I have been really struck by is how well integrated the Phantom 4 is. In Andrew Baker’s Phantom 4 teardown, you can see the integration polish in the Phantom 4.

The inside of a DJI Phantom 4 (teardown by Andrew Baker, used with permission from the photographer)

Less than 3 years ago, this is where we were (original DJI Phantom):

The DJI Zenmuse H3-2D installed requires the new NAZA PMU v2, so it gets a little crowded in the Phantom's shell. I also crammed an ImmersionRC 5.8Ghz 600mW transmitter in there (shown with antenna attached).
An original DJI Phantom modified to include a NAZA PMU v2, Zenmuse H3-2D gimbal controller, and ImmersionRC 5.8Ghz 600mW FPV video transmitter.

I modified this particular Phantom to include a gimbal controller and power management unit as well as a FPV video transmitter, and in addition to those parts, you can also see what are essentially multirotor kit parts crammed into a plastic shell. The flight controller, compass/GPS receiver, ESCs, motors, etc. are all clearly visible and identifiable, and do not look much different than the parts you might have found in a DIY quadcopter kit at the time. The components in each successive iteration of the Phantom have become more and more integrated, and there is no reason to suspect that this pattern will change, going forward. DJI drones have become all-in-one units, and with the latest drone, adding third party hardware has become almost impossible. Some folks complain about this, but you also don’t see anyone opening up a mobile phone and trying to add hardware to it.

The relentless drive to develop and integrate new technology is one of the things that makes DJI a tough company to compete with. Integration is in its blood, and being headquartered in Shenzhen, China, gives DJI the ability to tremendously shorten development timelines.

This article isn’t really about DJI’s prowess—it’s really about systems integration becoming required for the success of any new drone player looking to enter the industry (or survive, going forward). A few of the new drone companies I’ve seen actually do plan to release well integrated products, but there are many more who are banking on a single novel feature without having spent much time figuring out where the features will live. It’s certainly likely that some companies that have developed an amazing new technology will succeed, but any company that wants to make and sell drones shouldn’t underestimate the amount of effort it will take to actually create and manufacture a polished product, especially as commodity “drone-on-a-chip” clones become more and more feature rich, and as drone platforms like Qualcomm’s Snapdragon Flight threaten to push a lot of more interesting drone development into the software realm.

Here’s an example of a recent announcement that made me think about systems integration in drones. Skydio released a new video two days ago that demonstrates what their CV-based navigation system can do:


Skydio‘s compelling prior research video, which contains research from the founders’ MIT days, shows functioning SLAM and aggressive navigation from map data. The folks at Skydio have been developing this technology for years, and it’s likely that they have best-in-class vision systems and autonomous navigation algorithms for drones. But let’s look at the actual drone shown in the video:

Skydio's demo drone for CV-based navigation
Skydio’s demo drone for CV-based navigation

The demo is compelling from a navigation-as-a-feature standpoint, but the demo drone is really rough. If Skydio doesn’t plan to build and sell a drone, this isn’t a problem, but if they do plan to release an actual drone as a product, especially in the consumer space, this video could be an indicator that they are still at least 9-12 months away from announcing a product (of course, Skydio could also just not want to show the public where in the process they really are, or they could have other plans for a product or service). As a point of reference, 9-12 months is 1 product release cycle for DJI, and we’ve seen that DJI products do not necessarily only get better incrementally in each cycle—there are often big leaps in technology. I have high hopes for Skydio because I’ve met a lot of their early team and know that they are extremely smart, but the state of their publicly-shown demo drone has me a bit worried.

In the last 3 years, drones have evolved from kit parts in a shell to well integrated systems. It’s been an incredible thing to see happen, and I’m really excited to see what happens in the years to come!

Original Article