Rabitə və İnformasiya Texnologiyaları Nazirliyinin elektron xəbər xidməti
Skydio’s AI-powered autonomous R1 drone follows you around in 4K
Skydio, an autonomous drone startup based out of Redwood City, California, unveiled a product today that it’s been working on for four years. The R1, as it’s called, is an artificial intelligence-powered quadcopter capable of shooting 4K video of a subject and maneuvering complex environments all on its own. Using industry advancements in AI to help train a custom computer vision system, Skydio developed a product that’s effectively the first professional-grade drone that can be flown without any expertise whatsoever, the company claims. In fact, the R1 doesn’t even come with a controller because in almost every situation imaginable, the drone should fly itself.
It’s a dream quite a few startups have tried and failed to deliver: full autonomous drone flight that is both safe and robust at high speeds and in busy environments. But few companies, beyond drone heavyweight DJI, have come even remotely close. Most drones today, including both cheap entry-level units and multithousand-dollar rigs designed for professional use, involve by default a manual system using an external controller, which relies on at least a bit of know-how to operate.
For low-cost DJI drones like the Spark and Mavic Air, the learning curve is accommodating and the drones are relatively easy to pilot manually, but the autonomous assist features must be turned on and a bulky external controller is still necessary at all times. Most drones also stabilize themselves using GPS, and many perform poorly without a stable satellite connection.
Skydio’s R1, on the other hand, uses 13 onboard cameras to do real-time mapping, path planning, and obstacle avoidance. A Nvidia Jetson TX1 computer, used mainly as the brain of prototype self-driving car systems, is used for onboard processing. Skydio’s software combines all that data with built-in algorithms trained to recognize humans and other distinct objects like trees and cars. That way, the drone is able to avoid dangerous elements in an environment and follow a user through a variety of terrains and situations, from skis slopes to mountain bike trails.
In a demo yesterday on the edge of John McLaren Park, in the Excelsior neighborhood of San Francisco, I saw the R1 in action — and it works as advertised. CEO Adam Bry pulled the R1 out of a backpack as if it were a MacBook Pro, and he had it lift off using the device’s mobile app. Within seconds, the R1 was dodging and darting through a maze of trees, with the drone adjusting its distance and height to account for Bry’s sporadic movements.
The R1 had never seen this particular environment, nor had it ever laid eyes on me, yet within seconds it was following me around as well as I tried to crash it into tree trunks and random branches. All Bry had to do was tap the image of me on the live feed in the Skydio mobile app, and the R1 picked up on my presence. Bry says that the R1 takes into account visual data — like the blue color of the jacket I was wearing — to improve its tracking in real time, allowing it to better keep track of subjects that are in dense, object-rich environments.
All controls for the R1 are managed via Skydio’s mobile app, which manages the launching and landing of the drone. The app is also where users can see video previews of footage and control how the drone captures that footage with presets like “orbit” for 360-degree shots and “lead” to make the drone follow and film its subject from the front. The app has a manual flying mode as well, but all of the drone’s autonomous safety features remain active at all times, so the drone can’t be flown into people, trees, or other obstacles. Controlling the R1 is pretty straightforward, with an on-screen vertical slider and a virtual directional pad for sending the drone forward and backward and for rotating it left or right.
The R1, which goes on sale today for $2,499 and ships in about two to three weeks, is the product of a roughly 60-person team headed up by former MIT graduates. Bry and Skydio CTO Abe Bachrach were the first full-time engineering hires for Google’s Project Wing, the X lab’s drone delivery initiative, where they applied their existing expertise in autonomous flight systems. Both worked on such systems in an extensive research capacity back in Boston. Bry says the duo, who created Skydio alongside fellow MIT grad and software interface specialist Matt Donohoe, became intimately aware of the challenges in designing such systems. Many of the issues he says hinged on the development of a sophisticated enough vision system for the drone to understand its environment and react accordingly.
“When we started working on this, even in 2014 when we started the company, the leading computer vision-based flight results at the time were super, super primitive. Very slow speed, very clunky, no real demonstration of obstacle avoidance,” Bry tells me. “We made a pretty big bet that computer vision was going to be the future of this stuff.” Through machine learning, which allows software to improve over time through a complex system of self-improvement and pattern recognition, vision-based algorithms have advanced significantly in recent years to the point where even consumer-grade smartphones are capable of performing complex object recognition and other human-like feats.
“Vision is just a super rich source of information and the algorithms are getting better and better at extracting that information,” Bry adds. “There are a number of areas where I think we’ve meaningfully advanced the state of the art. If we were a research lab, we probably would have published a bunch of papers. But I think we’re also just fortunate to be riding this really amazing trend in AI and computer vision.” For instance, Bry says that when Skydio was founded, a computer powerful enough to handle the type of local processing necessary for advanced autonomous flight did not exist. But Nvidia developed its Tegra-based TX1 computer for that express purpose, with devices like self-driving cars and drones in mind, leading to a fruitful partnership.
Skydio pictures the R1, which comes with about 16 minutes of flight time on a single charge and 64GB of onboard storage, appealing to the the early adopter crowd within the drone enthusiast community. But the company also sees the device appealing to anyone who’s interested in the type of photography and video drones are capable of capturing. The company, at least right now, has a specific focus on action sports enthusiasts and athletes, the same people who populate many of the company’s early testimonial videos. Because the R1 requires no expertise, it can be flown out of the box with all the sophistication and safety considerations of an experienced pilot, the company says. That should help it catch on with hardcore runners, skiers, and surfers as much as it does with experienced UAV pilots.
Founded in 2014, after Bry and Bachrach’s stint at Google, Skydio has amassed $70 million in funding from Andreessen Horowitz and Nvidia. (Golden State Warrior NBA star Kevin Durant also counts himself as investor as well.) Its most recent funding round — a $42 million Series B led by firms IVP and Andy Rubin’s Playground Global — is being announced today in conjunction with the R1 launch, and the money is poised to keep Skydio healthy enough while it produces a limited amount of R1 “Frontier Edition” units at its HQ in Redwood City. Bry said he could not disclose whether the company plans on moving production to China, or when a second round of R1 devices would go on sale. But it’s clear the company has serious momentum for the first batch of its consumer product debut.
The confidence of Skydio’s investors and the funding amount for a hardware drone startup is certainly notable. But there remains a level of skepticism necessary when talking about the viability of a high-end product in today’s drone market. For customers with the means to purchase a $2,500 gadget, there’s already proven and well-reviewed products from China’s DJI, far and away the world’s leading drone maker.
You could buy the $2,999 Inspire 2, which goes nearly three times is fast, lasts almost twice as long on a single charge, and shoots 5K video, among a dizzying number of other perks. You could also buy DJI’s Phantom 4 Pro for $1,499, which also beats the R1 in those metrics. Both drones come with DJI’s existing and constantly improving autonomous features, which include a GPS and infrared-powered sense and avoid system and a number of other sensors and software-powered advancements that let DJI products avoid obstacles, track subjects, and follow flight plans.
Skydio is confident that no product, not even a DJI Inspire 2, can compete on autonomy with the R1. From my admittedly brief demo, the company’s product does seem more sophisticated than anything we’ve seen in the past. The ultimate goal, of course, is not to compete with a high-end DJI drone, but to bring the level of autonomy the R1 delivers to a cheaper, more mainstream consumer product, much like how DJI has made traditional, manually controlled drones more accessible with devices like the Spark and Mavic Air. Bry compares the R1 to a Tesla Roadster, in that it packs in a bunch of forward-looking tech into an enthusiast product with a high price tag.
More traditional devices, Bry thinks, will always be marketed toward UAV junkies and the types of consumers and professionals who enjoy piloting aircraft. For everyone else, there’s not a great use case for a drone. The R1, however, is not about piloting a 4K camera, but rather what such a device lets you do, specifically when you remove all the traditional hurdles involved with piloting one.
“This product is less about the experience of manually flying it and much more about the content that you can create. There are always going to be people who enjoy flying, which is great,” Bry says. “But our bet is that there are a lot of people who are excited about what a drone can do, but who are more excited about what it can do for them.”