If you’re sick of selfie sticks, Boston-based software company Neurala may have an alternative for you. The Selfie Dronie is a paid mobile application compatible with Parrot Bebop drones that offers users a relatively hands free way to record selfies and dronies (those aerial shots often associated with extreme sports and Redbull advertisements).
“It’s honestly a fun marketing stunt,” says Roger Matus, VP of Products and Markets at Neurala. He isn’t downplaying his company’s app’s innovation. Rather, Matus is admitting that Neurala’s gaze is set on loftier pursuits than the perfect selfie.
To understand Neurala’s end goal, it helps to understand how Selfie Dronie compares to other auto-follow drones and drone footage apps on the market. “Tethered” apps like AirDog and Hexoplus use GPS to facilitate tracking. For AirDog, that tracking is based on a wristband, which communicates relative position to an airborne drone, making it valuable for extreme sports and nautical activities. Meanwhile, Hexoplus connects to a user’s smartphone, transmitting GPS positions back and forth to stay within range. Of course, the GPS reliance of these apps leaves a margin for error as both the drone and the wristband/smartphone transmit a relative position.
Neurala on the other hand uses machine vision to track and follow users. After instructing the drone to take off, users draw a box around the subjects they want included in the footage. The drone then hovers from side to side (for a selfie) or flies around at angles (for a dronie), depending on which option the user selected. Matus admits the technology is “very successful at taking selfies and dronies, but not a whole lot else.”
The key here is the vision, which Neurala wants to apply beyond just drones. The Selfie Dronie app is really an attempt to display the company’s neural parallel processing technology.
Neurala was created when its founder predicted that GPUs and massive parallel processing would be a way of emulating the human brain and performing artificial intelligence. Matus explains: “Graphic processing units are very good at doing a doing one thing a lot of times. They started in screens, hence the term ‘graphics’. If you have a 1024×768 screen, that’s 1024×768 (786432) operations happening simultaneously. It’s the same operation but there’s a lot of them. That’s basically parallel processing.” The company now holds the patent for GPU-based systems running artificial neural networks. When Neurala filed the patent for such technology in 2006, Matus says people thought they were crazy. “Now what we’re seeing this year is a lot of applications and a lot of hardware are stating to incorporate GPU processing into devices in order to do decision making.” Matus predicts that by the end of this year, toys will be equipped with these processors so they can make decisions. “We’re releasing a whole bunch of software to enable that to happen,” he says.
Indeed, by displaying Neurala’s GPU and massive parallel processing technology through an app that’s mostly geared towards consumers, the company hopes to attract attention from toy manufacturers and creators of consumer electronics. “We want to help them make their toys more engaging,” he says, “and to play in the world in which the owner lives.” Beyond toys even, Neurala hopes to introduce their software to self-driving cars to “anticipate problems before they happen.” Matus says, a self-driving car may learn to stop when a person jumps into the road, “but we want to use vision to see people on the side of the road who are about to go in front of the car. And we want to anticipate, not where people are, but where people are going to be.”
Matus calls this “actionable perception,” or a machine’s ability to perceive the world as it will be and take action. “These cute software products (like the Selfie Dronie) are showing this advanced AI technology that will start permeating a lot of spaces.”
