A drone that can dodge, duck, dip, dive – AOPA Pilot

Photo by Terry Jarrell.

Many drone designers have tried to automate obstacle avoidance for obvious reasons. Yuneec turned heads at the 2016 Consumer Electronics Show with the Typhoon H, demonstrating on-board obstacle avoidance technology to follow a mountain bike through an obstacle course made of fake trees and rocks. Market giant DJI countered with its own obstacle avoidance automation. While these early efforts were somewhat effective, they had limits.

Robotics engineers took another step toward true autonomy with the original version of Skydio. It was pretty good at following a subject and dodging obstacles, but everything else about this drone was underwhelming. A limited control interface, camera performance that did not measure up to the competition, and short flight time made it more of a curiosity than anything. Did I mention it was expensive?

The Skydio team knew what they were on to, though, and already had Skydio 2 on the drawing board. In case you haven’t heard, this has become the “it” drone of the moment, setting the internet ablaze with chatter, curiosity, and demand. This iteration of the Skydio drone comes with a respectable camera, able to deliver 4K video at 60 frames per second.

Cameras are also an integral part of Skydio’s superpower: the ability to follow a moving object through a complex environment at a very respectable speed. Skydio 2 sports six high-definition cameras dedicated to constantly scanning subjects and obstacles, and feeding 360 degrees of artificial situational awareness into a multicore computer programmed to build a 3D model of the drone’s surroundings in real time that the Skydio 2 can then use to navigate around and avoid fixed obstacles while tracking or following a subject.

The obvious benefit of this setup is that it lets the operator focus more on their activity—such as biking, skiing, or whatever they are doing—and leave the dodging to the drone. The worry factor of slamming into a tree or building is greatly reduced by all this computer horsepower. This really is the next-generation selfie stick we’ve been waiting for.

Complacency not an option

I have heard the phrase “they fly themselves” more times than I care to count. Many inexperienced drone operators confuse the built-in flight modes such as orbit, rocket, dronie, and others with true autonomy. While these preprogrammed modes sometimes make it easier to get smooth shots or effects, they can have lapses when it comes to executing the shot and dodging the tree. Automated flight can increase the risk of a collision, particularly if winds are gusting, or EMS helicopters arrive in the area unexpectedly.

As a remote pilot, you, and only you, are responsible for what your drone does at all times.

This brings up an interesting question, specifically about the potential to fly beyond visual line of sight (BVLOS). Much of the marketing and online hype for Skydio 2 shows videos of the drone tracking people doing various activities such as biking, skateboarding, skiing downhill, and similarly dynamic activities. They show that once the drone is launched, you can take a “set it and forget it” approach and go about your business zooming down the road on your motorbike with your ever-so-faithful Skydio 2 humming along behind you.

I spoke with Ken Dono, a highly popular and informative blogger on the latest drone models at OriginaldoBo on YouTube, about his hands-on experience with his Skydio 2. Ken pointed out some of the differences in flying this new type of flight-assisted vehicle.

“Many times, it just feels like someone else is flying the drone,” Dono said. The AI wants to make decisions that may differ from what the remote pilot has in mind. As is to be expected with this sort of new technology, early adopters are “beta testers, finding some weaknesses and limitations.”

For example, the initial rollout of the Skydio 2 did not even provide telemetry information to the pilot, and the controller is borrowed from the Parrot Anafi with Skydio’s software baked in.

Despite these early hiccups, Dono said the Skydio 2 shows great promise: “I see this tech moving into commercial and even BVLOS applications” as the technologies continue to evolve and improve.

So, what happens in the event of an emergency?

Picture this: You are busy riding your bike in this scenario, not even looking back at a situation that the drone is about to encounter. No drone is impervious to fly-aways, and unexpected encounters are pretty common, along with any number of other surprises that could cause a dangerous situation that, in the worst case, leads to property damage or injury.

Dono echoes my earlier comment about not getting complacent. “Great power comes with great responsibility. It is important to not have a false sense of confidence” and put all your reliance in the drone’s control.

That is easier said than done if you’re racing down a mountain on a bike.

First, but not the last

While Skydio 2 may be the closest approximation of “true” AI in the drone market yet, the competition is hot on Skydio’s heels.

The internet is starting to buzz about the upcoming update from Autel to its Evo line with the Evo 2, and this thing looks to be packed with sensors all around. While the Evo 2 has just been formally announced and shown at CES, it appears to have some variation of advanced, 360-degree obstacle avoidance. Some highlights for this model include advanced 6K and even 8K interchangeable cameras, which obviously raises the bar for professional and semi pro photographers and videographers. There is even a dual camera setup for commercial applications that looks amazing.

DJI should be updating the Mavic 2 Pro and Zoom models, thought the company is tight-lipped about development projects. It would not be surprising to see an updated Mavic 3 boasting more of these forward-looking features.

AI and better flight autonomy sure do look to be an incredible new layer to the drone technologies that we are already using and enjoying. Avoiding trees and other close-range obstacles is one thing they can do better than a human, if not now then very soon. But thinking things through ahead of the flight, anticipating birds and other unexpected encounters, is a human trait that computers won’t beat us at any time soon. The key is understanding that these new tools are complementary to real piloting and not a replacement for human skills and judgment.