The other day, my coworker was showing off his new DJI Spark drone. A few of us hiked up a nearby trail. When we got to a nice viewpoint, I watched as he held the small device in the palm of his outstretched hand and directed it to take off. We waved as the drone slowly climbed upward, recording video as it rose. Then he demonstrated how to take a drone-powered selfie by using his hand to direct the drone to fly a few feet away and fire off a photo of the group.

The DJI Spark drone dutifully records video of us and the surrounding landscape.

Later, I started wondering how so much intelligence could be packed into such a small, autonomous device. That’s when I connected the dots between several blog posts I wrote in the last few months. In one story, I talked about modern uses for commercial drones in various industries. In other posts, I talked about artificial intelligence (often known as AI) and showed some real-world examples of AI. I admit that I didn’t connect drones and AI in my mind when I wrote those stories, but the two technology areas are inextricably linked.

Without recent, rapid advances in AI, my coworker’s drone wouldn’t be able to do any of the incredible feats he showed me. Instead, my colleague would be struggling to manually navigate the device around obstacles by using a cumbersome remote control with a joystick. Even safe landings would be challenging without practice, because he’d need to manually control speed during descent, while taking into account wind and other factors. AI software and hardware automatically handle all those factors—power, wind speed, direction, acceleration, altitude, and so on. It’s AI technology that separates modern-day autonomous drones from the large, expensive, remote-controlled planes and helicopters of my youth.

The DJI Spark mini drone

I was curious how DJI—the maker of the Spark drone—packs so much intelligence and processing power into such a small, light package. After all, the Spark drone only weighs 300 grams (about 10.5 ounces). I discovered that the drone has an array of cameras and sensors feeding data into a tiny chip called a vision processing unit (VPU) that’s less than 6.5 square millimeters. That tiny chip, called a Myriad 2 VPU, is made by Movidius, an Intel subsidiary. The VPU is a type of microprocessor that’s designed to accelerate machine vision tasks, such as converting visual signals into data that can be used to steer the drone. In the case of the Myriad 2 VPU, imaging, computer vision processing, and something the company calls a Neural Compute Engine are all combined and managed on that one, small chip.

With a little more digging, I learned that the Neural Compute Engine is used in a wide range of technologies, including smartphone and tablet cameras, action cams, electronic eyewear, home automation devices, and industrial robotics. All of these devices use the engine to process massive amounts of sensor data at high speeds, in order to allow nearly instantaneous actions based on that data. For example, our smartphone cameras can rapidly analyze a scene to distinguish distances, faces, and backgrounds and adjust based on subjects, lighting, and other factors in order to take the perfect selfie or cat photo. Industries can use similar technology to process incoming data on sensitive machines in factories or to monitor conditions on delivery trucks, for example.The Neural Compute Engine is an on-chip deep neural network accelerator that allows the VPU to process more than two trillion 16-bit operations per second.[1] What that means is that the VPU can process visual inputs and make intelligent decisions based on commands and its surroundings, instantaneously, without the time and power constraints of communicating with the cloud. Everything is handled autonomously on the device itself.

Companies like Intel and Movidius are constantly pushing the boundaries of what’s possible with AI. The tiny VPU chips provide deep learning capabilities far beyond what desktop systems offered just a few years ago. And with each passing year, these chips are getting more powerful while their dimensions and power consumption shrink. In fact, only a few weeks ago, Movidius announced the Myriad X chip, which features an all-new Neural Compute Engine designed to dramatically increase performance of deep neural networks. According to Movidius, the Myriad X delivers 10 times the performance of previous-generation VPUs.[2]

At the rate things are changing, I wouldn’t be surprised to see a pocket-sized drone on the market within the next few years. In the meantime, you can learn more about AI and drones at the Intel iQ site. And to keep up on the latest developments in AI and other technologies, follow Prowess on our blog, Twitter, and LinkedIn.

[1] Movidius. “Myriad 2 Vision Processor.” 2014. http://uploads.movidius.com/1441734401-Myriad-2-product-brief.pdf.

[2] Movidius. “On-Device AI and Computer Vision. Myriad X: Ultimate Performance at Ultra-Low Power.” www.movidius.com/myriadx.

Share this:

FacebooktwitterlinkedinmailFacebooktwitterlinkedinmail