Imagine turning your day into an action film or being able to monitor your children as they make their way to school or to a friend’s house—all with minimal effort. The best part of this is you don’t have to imagine it. It’s real, and it’s happening all around us right this second.
Autonomous drone technology has reached the point where this, and much more, is possible. You can bike through urban centres or through wooded parks and have a drone follow you with almost no input on your part. These devices can navigate with remarkable precision, following you, all while never missing a beat. You set it and forget it, at least as long as the battery life allows.
This is technology that gives more people the dynamic filming quality once reserved for professionals. The cost of entry is dropping, and they’re more user-friendly than ever before. If you’re an aspiring videographer—or you just want to take exceptional video—you need an autonomous drone buzzing by your side (or above your head).
Image courtesy of Pixabay.
Ultimate Action, Ultimate Control
While autonomous, self-driving cars have made headlines around the world for the past few years, autonomous drones are just entering the mainstream. Autonomous cars and drones share much of the same technology, at least when it comes to their decision-making artificial intelligence. The same processes that tell a self-driving car not to hit that sign also tell the drone not to hit a tree.
There’s a big reason why autonomous drones are only now gaining popularity, even though many models have been available for the past three to four years. Self-driving cars operate in a two-dimensional space (or 2.5-D space). Autonomous drones must function in a fully three-dimensional environment. Avoiding one tree is complicated enough. Avoiding every tree in the forest takes serious processing. Technology is finally catching up with our big ideas.
Many companies have attempted to create autonomous drones that could navigate through just about any environment. This includes China-based DJI, one of the bigger drone manufacturers in the market today.
Another company, Skydio, an upstart out of California, has also thrown their hat into the autonomous drone arena. Since releasing their drone—the R1—earlier this year, they’ve made quite the impression. The R1 has become the drone people want. If you can get your hands on one (it’s not yet readily available outside of the US and Canada), it’s worth it for many people.
The Challenge of Adaptive Technology
Skydio wanted to create a drone that could fly itself wherever it was needed by a user. The R1 connects to a smartphone, you tell it what to do, and it does it, for the most part. There’s no remote or regular input on the part of the user. To accomplish this, the company developed artificial intelligence software, not unlike the AI used by self-driving cars.
Coupled with this AI, Skydio’s R1 drone packs 13 small cameras that work independently of the main mounted camera. These mini cams allow it to map its immediate surroundings in real-time, giving the AI the information, it needs to navigate without clipping or hitting any obstacles.
For processing, the R1 is equipped with an NVIDIA Jetson TX1 module, a relatively low-power processing unit designed for onboard devices, which includes self-driving cars. The brains of the TX1 is a Quad ARM A57/2MB L2 processor, along with 4GB 64-bit LPDDR4 25.6 GB/s onboard memory. You can see the full array of technical specs on NVIDIA’s website here—which includes the specs of the Jetson TX2, the latest generation module.
Image courtesy of Pixabay.
Phenomenal Processing Powers, Itty Bitty Physical Space
The TX1 may be a “last-gen” processor (it is nearly 3 years old), but we appreciate this cost-cutting measure. The devs seem to have a handle on what is needed in their device and by using the TX1, rather than the TX2, the savings are passed onto the consumer. Even better, though, development for the TX1 is further along than the TX2 and since 2015, communities of devs and enthusiasts dedicated to the TX1 have popped up online.
For hobbyists, makers, and developers, TX1 modules can be purchased through many major online retailers. The kits offer a great opportunity to experiment with the technology. You get access to a mini supercomputer (as its advertised) that is said to have more processing capability over the current Intel Core i7 CPUs, while not taking up much more space than a deck of cards. That, at a glance, is impressive. Cutting through the marketing-speak, it is still an older module that may not hold up to real-world scrutiny.
Thanks to its machine learning capabilities, and with the right software, the Jetson TX1 can still accomplish many remarkable tasks—as Skydio put to the test. The company successfully trained their R1 drone to distinguish individual humans from other subjects, such as other people, trees, and animals. This training allows the R1 to seamlessly follow a person in just about any environment, while safely avoiding other objects and people.
Since its release earlier in 2018, the R1 has proven its capabilities are well thought out. The technology works, though it’s not perfect. To be “perfect,” it still needs to master night-flying and boost its all-weather ruggedness. And then there’s the question of longevity. That is to say, battery life.
The R1 Vs. the Inspire 2
Between flight, filming, and data processing, users of the R1 have 16 minutes of sustained flight time before the onboard battery must be recharged. The device does ship with a second battery, so you don’t have to worry about charging immediately, but for a total of 32 minutes of in airtime, it’s likely you’ll be scrambling for a power point.
It’s a major downside, for sure, but until lithium-ion battery technology improves, or new battery tech hits the market, it’s a downside we just have to live with. For the average consumer, it might even mean waiting for the next-gen version of the device to see how the battery-life issues are ironed out.
A similar drone, the DJI Inspire 2, packs 27 minutes of flight time, which is comparatively impressive. This feat is large in part to its lightweight magnesium-aluminium alloy body. The R1 has an aluminium body with carbon fibre rotor guards.
The DJI Inspire 2 does have a few other advantages over the R1, as well. At the moment, the R1 films in up top 4K. The Inspire 2 has a video processing system allowing for 4K, 5K, and 6K video. The Inspire 2 can also travel at up to 58 miles per hour, though for the obstacle avoidance technology to work accurately, it cannot be travelling more than 45 miles per hour. The R1 tops out at 25 miles per hour.
Image courtesy of Pixabay.
The Future Is Wild
These two autonomous drones are just a small sample of how far drone technology—as well as machine learning—has come. It’s a glimpse into something bigger, from self-driving cars to drones that may very well become our daily companions. It’s an opportunity, too, for a lot of people to get in on the ground floor, so to speak, of the autonomous drone revolution. You can help develop new capabilities for these devices and pave the way forward for the next generation of these incredible machines.
Even now in the United States, Shawn Loftin of Helio RC Innovations is optimizing his own Inertial Measurement Unit Filter (IMUF) system for Helio’s drones.
“We designed a coprocessor that runs in parallel to filter a gyroscope,” Loftin said. “The IMUF system uses a ICM-20601 and STM32F301 and delivers filtered gyroscope data at 32 kHz or delivers quaternion for flight at 16 kHz. The IMUF system is easily added to existing hardware designs to decrease cost of components, measurement error rates, and latency of data. The IMUF system includes a specialized Kalman Filter that adapts to gyroscope noise in real time using machine learning to ‘tune’ the filter during flight.”
Loftin and Helio added this system to their flight controller hardware, “which reduced the load of the main MCU so much we doubled the controller loop frequency. This also allowed us to add more features for customers to use in flight while creating a noticeably more stable and predictable flight with our advanced filtering. The coprocessor design helped us avoid the time cost of developing firmware for a faster and more expensive processor which would still have more latency with robust filtering.”
With further innovation like this, in 10 years, homes across Britain and beyond will very likely be equipped with the autonomous drones, fetching the post, walking our dogs, filming our daughter’s wedding, or restoring a power line after a windstorm. This a list that can go on and on. By the end of this year alone, there’s going to be a lot of remarkable change, and innovative engineers are going to spark a lot of it.
