Movidius, a company that’s building low-power graphics processing chips for computer vision, has announced its next-generation processor that is 20 times faster and consumes less than half a watt of power. Movidius may not be a household name, but it has a very big supporter in Google, which is using the first-generation Myriad chip from the startup for its cell phone vision project known as Project Tango.
A Movidius chip acts as the eyes of a smart phone for Project Tango, which my colleague Kevin Tofel has described a few months back as a precursor not just for virtual reality on a handset, but a key to cheaper robots. He wrote:
In fact, I see Tango as a better VR option than Google Glass. Glass excels at providing useful, contextual information outside your field of vision while Tango is likely better suited for your full attention. Or the attention of others; and by others, I mean robots. The consumer robotics industry could see the biggest benefit from Tango than any other space if it the technology works and is cheap enough. After all, you can’t have robots with mobility if they can’t get around in a world they can’t see or understand.
In a call with Movidius CEO Remi El-Ouazzane, he discussed the system on a chip design of the Myriad2 processor, including the need for programmers to run a specialized compiler to on software written for the chip. That’s always a bit of a challenge in getting developers to build applications for silicon, but with Google leading the way that could be overcome. Another interesting aspect about the Myriad2 is that it works as a co-processor with an application processor on a cell phone, or could be paired with a much simpler microcontroller for a device like a fitness tracker or sensor.
That means you could essentially give vision of sorts to a variety of really small, dumb devices. They’d have to have a way to ship images to the cloud since many of these devices don’t have much screen real estate, but it’s fun to think what you might do with a fitness tracker that knows that you’re walking in the city as opposed to on a trail. Or a smartwatch that can “see” your environment and use that to derive context so it can show you only relevant notifications.
El-Ouazzane told me in an interview that the system on a chip will cost in the “single-digits” depending on how tricked out the processor is, which is a bit on the high side for low-end sensors but something one could see in a high-end smart phone as a differentiating feature. And over time, Movidius could benefit from the economies of scale that cause electronics to get cheaper as more devices are produced. So let the age of computer vision begin.