A Parrot AR Drone controlled with an NVIDIA SHIELD handheld, Credit: Parrot
NVIDIA (NVDA -3.33%) is best known for its high-end graphics cards, but it's no mystery the company has broader ambitions.
We already know, for example, NVIDIA's Tesla GPUs power the world's 10 most energy efficient supercomputers. And with the launch of products like the NVIDIA SHIELD handheld, tablet, and gaming controller, as well as NVIDIA's cloud-based GRID gaming service, it's striving to build awareness for its long-term vision of low latency, console independent gaming.
That's also not to mention NVIDIA's efforts to take the automobile industry by storm, providing chips to not only power infotainment systems, but also to handle the computationally intensive imaging tasks required to enable self-driving vehicles.
But cars aren't the only thing NVIDIA is helping become more autonomous.
Driving the future of drones
To be sure, NVIDIA recently highlighted its role in the work of up-and-coming drone maker Parrot. Since bursting onto the scene five years ago with its mass-market-oriented, smartphone-controlled AR Drones, Parrot regularly awes onlookers with its slick demos. This year at CES, it even showed off its new minidrone line with a mesmerizing choreographed dance of sorts.
But arguably Parrot's most impressive CES demo, called "Kalamos," was constrained to a small table and involved no flight at all:
Parrot's "Kalamos" demo is powered by an NVIDIA K1 processor. Credit: NVIDIA
Instead, Kalamos was comprised of a robotic arm that mimics the motion of a next-generation drone with dual cameras and sophisticated computer vision capabilities. If you look closely, the latter is no surprise considering an NVIDIA logo is contained underneath its transparent cover.
To be sure, Parrot utilizes an NVIDIA Jetson TK1 development kit, which itself contains a powerful 192-core NVIDIA Tegra K1 GPU. What's more, Parrot uses an NVIDIA SHIELD controller to direct the "drone," whose cameras act much like human eyes in the way they can process stereoscopic images to create a 3D model of whatever they see in real time.
Think of it this way: Our eyes are spaced slightly apart, so our brains are able to process the slight differences in the angle at which the two images are collected. By combining those images, our brains then translate it to an understanding of depth. With Kalamos, Parrot is programming its drones to do exactly the same thing. Sounds simple enough, but it's an extraordinary engineering feat that requires massive parallel processing power to perform in real time -- something to which NVIDIA's multicore GPUs are uniquely suited.
In this case, Kalamos' test subject happens to be the toy house, the surfaces of which were reconstructed for display on a nearby tablet as the robotic arm moved around it. So rather than relying on traditional GPS or rudimentary object detection to navigate as many past drones do, Parrot's Kalamos leverages the incredible computing power and parallel processing capabilities of NVIDIA's Tegra K1 to actually "see" and model the world around it in real time.
Of course, it also helps that NVIDIA designed the K1 with energy efficiency and mobile applications in mind; Battery considerations are paramount to any processor's ability to be incorporated into a small drone.
This is just the beginning
But we should also keep in mind only a few weeks ago, NVIDIA also unveiled its new 256 GPU core Tegra X1 superchip. According to NVIDIA, the Tegra X1 offers twice the performance of the Tegra K1 on every metric, and marks the first mobile processor to achieve over one full teraflops of computing power. If the folks at Parrot can use the K1 to effectively build real-time models of a drone's surrounding environment, I can't wait to see what they'll do once they get a Tegra X1 dev kit in their hands.
So, how big is the market for these drones? According to aerospace analysts at Teal Group, sales of unmanned aerial vehicles are expected to nearly double over the next decade to around $11.5 billion, including around $1.6 billion from civil sources. That might not seem like much as NVIDIA's take as a chip supplier will be a fraction of that amount, but keep in mind drone makers to date haven't had much motivation to adopt high-performance GPUs in their products.
Over the long term, as Parrot continues to demonstrate the possibilities for using GPUs in next-gen navigation, the drone market could ultimately represent a solid incremental win for NVIDIA.
