Intel’s Vision For Single Board Computers Is To Have Better Vision

At the Bay Area Maker Faire last weekend, Intel was showing off a couple of sexy newcomers in the Single Board Computer (SBC) market. It’s easy to get trapped into thinking that SBCs are all about simple boards with a double-digit price tag like the Raspberry Pi. How can you compete with a $35 computer that has a huge market share and a gigantic community? You compete by appealing to a crowd not satisfied with these entry-level SBCs, and for that Intel appears to be targeting a much higher-end audience that needs computer vision along with the speed and horsepower to do something meaningful with it.

I caught up with Intel’s “Maker Czar”, Jay Melican, at Maker Faire Bay Area last weekend. A year ago, it was a Nintendo Power Glove controlled quadcopter that caught my eye. This year I only had eyes for the two new computing modules on offer, the Joule and the Euclid. They both focus on connecting powerful processors to high-resolution cameras and using a full-blown Linux operating system for the image processing. But it feels like the Joule is meant more for your average hardware hacker, and the Euclid for software engineers who are pointing their skills at robots but don’t want to get bogged down in first-principles of hardware. Before you rage about this in the comments, let me explain.


This is the Euclid. Its size and shape reminds me of digital recorders from the 90’s and early 2000’s, but it has a sleek glossy black finish (a bit different from when first teased in August) and it’s bristling with ports, buttons, and a few obvious sets of optics.

This beast is running a quad-core Atom processor with 4 GB of RAM and 32 GB on-board storage. That alone isn’t going to blow you away, but the Euclid also has a RealSense depth camera, an RGB camera, and a fisheye camera built in. It’s capable of stereoscopic vision (VGA resolution) and includes a bevy of sensors crucial for robotics like IMU and GPS. The thing even comes with a Lithium battery.

It’s running a full desktop install of Ubuntu which may sound a bit like overkill, but I think the whole point of something like this is for people who don’t want to learn a new platform. Next week we’ll be publishing an article about a group of hackers who were showing off their Autonomous RC Vehicles at the Faire. They have competed in Sparkfun’s AVC and slapped a Macbook Pro onto and RC chassis for the purpose. The Euclid is going to provide much of the same functionality they had for that build at between a fifth and a tenth of the price — all wrapped up in a cozy little case (standard tripod mount for an easy interface). Grab your wireless keyboard and plug into the HDMI to program and debug, or use the built-in WiFi to tunnel in.

In the case of the small wheeled robot you see here, the connectivity to the bot is accomplished through USB. A lower-level embedded board drives the motor controller, with serial commands issued from the Euclid. Some might criticize the latency of using USB, but SBCs controlling robots almost always have similar latency issues.


If you’re more into the bare PCB, the Joule which was announced back in August) is worth a look. It uses an approach now familiar from Intel’s “maker” offerings; Joule itself is a module that needs a host board to break out all of the connections. It is also running an Atom processor with 4GB of RAM and 32 GB of onboard storage. Sounds a lot like the Euclid, right? They’re close, but the two do have different processors.

Basically, Joule is the brains, and Euclid is a flashy way to market them… and like I said earlier, get more people hacking without getting stuck trying to connect bits of hardware together. But many of us love to connect bits of hardware together and that’s why the 48 broken out GPIO are a robot builder’s delight.

This module delivers the same ability to process computer vision as the Euclid, but you’ll need to bring your own cameras to the party. You could buy the Intel RealSense depth camera the Euclid is packing, but at the Intel booth, it was an underwater ROV that caught my interest.

The team at Rajida sourced a binocular camera board from TaoBao and built their ROV to follow colorful fish around. I got a pretty neat demo using a yellow angelfish toy on a stick. Unfortunately, the Intel booth was outside and sunny California is no place to record video of television playback. Enjoy the still images and trust me that this works and it’s cool.

Brian Benchoff often writes (ironically – ed.)  about the “selfie drone” being the killer app for portable computer vision. I’m thinking there would be a huge market for underwater ROVs that follow scuba divers and record the experience from a 3rd person perspective.

You can get your hands on a Joule now for $299-$349. Looks like Euclid isn’t shipping yet but pre-orders on Intel’s site start at about $399.

33 thoughts on “Intel’s Vision For Single Board Computers Is To Have Better Vision

  1. Joule is $299-$349. Euclid is $399. And that’s the sound of a limited market share. Sell these at $100 and you are starting to establish something. As it is, these are vastly overpriced. How open are they anyway?

    1. Closest non-stereo vision setup w/ dia range detection ( ) would be the combination of the jevois which has a camera ; arduino and modified laser pointer. ~$100. To do stereo, could add a 2nd jevois, but would need a much beefier board than the arduino.
      Not sure how much more processing one would need, as jevois does the heavy, mono-vision work.
      So the Joule isn’t to overpriced if one is doing stereoscopic vision.

    2. $399 is still probably a solid loss considering the low volume and amount of components included. And the guys that cannot even afford that most likely isn’t all that interesting to Intel anyways, so why throw more cash after them?

      1. I personally think it’s a mistake for Intel to ignore the Arduino / RPi / etc market and to be fair, it is not as if they are completely ignoring it; they just keep completely screwing up any reasonable attempt at hitting most of the sweet spots of what the market wants.

    3. I like how Intel makes these powerful PCs in a small package. However, I kinda wish they would just publish the algorithm for SLAM they are using as a free open source C library along with a research paper on how everything works. For example, like the AprilTags algorithm. Their platform would then be a “reference model” that everyone would use to test out code before building their own custom system using the algorithm.

      Maybe they do the above… but, it’s not presented in the media this way. For example, when Google released details about their Tensor Processing Unit that was done in a very open way.

    4. Performance of Joule and Euclid blows away any Raspi + Camera and OpenCV. You get what you pay for. Every part has its place. Neither is better than the other, just that some can do more, so you pay more. Joule is super small. Raspi has cajillions apps for it. I use both for different applications.

  2. “software engineers who are pointing their skills at robots but don’t want to get bogged down in first-principles of hardware”

    In other words, software engineers who can’t be bothered to learn the first principles of a field they’re meant to be pointing their skills at?? Good luck to them…

    1. I write ASP.Net apps in C#. 99% of the time, I don’t have to worry about how the Common Language Runtime actually does garbage collection, or how the Windows networking stack works.

      Most Arduino users don’t need to know how the toolchain works, what avrdude actually is, or even how the Arduino bootloader works.

      In this case, Euclid is a well-defined off-the-shelf target platform for developing a certain class of application.

      You’re several decades beyond developers actually needing to know the extremely low-level details of computer hardware. We’re not Grace Hopper hunting down actual bugs.

        1. Yeah but that’s the stilt-manufacturer’s problem. If one has metaphorical woodworm, they send you an improved one.

          To save torturing a metaphor, we can no longer be experts in every aspect of computers. There’s too much stuff to learn. Most programmers work in a certain field, and their time and brains are better spent improving in that field, than learning stuff that’s somebody else’s job. It’s gone from being a job for generalists, to specialists. Same as car design, same as any mature post-industrial field.

          Leave security to security professionals. There’s certain precautions we all need to take, based on the advice of those pros, sure. But having every programmer be their own security expert, is going to procuce worse programmers and worse security experts. Instead, we pay companies to do it for us, from the money we make being experts in our field. This is how all technologies progress. It’s how everything’s progressed since the invention of trade and specialisation, some time back in the Neolithic.

          1. I agree with your statements regarding security professionals/programmers.
            I don’t do much for home Internet security because I’m afraid to do the wrong thing and make things worse, or not enough. I do follow some general recommendations about router/firewall setup, but there are so many places to research, and which is the best one, and it all takes precious time. Programmers should be aware of practices to prevent stack overflows and other security holes, but they cant use all their time learning them or coding them.

          2. Yes it is cheaper, but there is one problem with using that approach and that is that a massive number of products with the exact same attack surface, if you can own one device you can probably own them all. Because their foundation would be exactly the same. On paper having a very small number of companies providing all security should be fantastic, global experts in their field, but product release deadlines, can mean that less than optimal decisions are made (not always but sometimes).

      1. That sounds like saying “Using the ‘Euclidbike®’ or the ‘Ardubike’ is so intuitive, you don’t even need to know where the brakes are”. It kinda works until something goes wrong, and according to Murphy’s law, you won’t have to wait long until something goes wrong.

        BTW I have nothing against user friendly stuff, but I’m still surprised how people do not try to understand it and just treat their tools as black-boxes.

        1. The idea of black boxes is why we have everything around us! Specialisation, and some certification to ensure the experts we pay to design jet engines actually know what they’re doing. That’s why very complex things like cars, aeroplanes, computers, are affordable. Henry Ford ripped off the guy who figured that out (can’t remember who invented mass production, but Ford has the fame for nicking the idea). Instead of a factory full of guys all making their own cars, you have one guy who knows how to do brake linings, and that’s all he needs to know. It increased productivity and made cars affordable to the masses.

          There’s only so many brain cells to go around, specialising in one area each is the most efficient use of them. I’m surprised anyone could think differently.

        2. This. It’s a challenge to balance everything and every user is going to have a different perspective who has a different perspective from a manufacturer or distributor, etc. There simply isn’t time to know everything about all of the details either. Almost nobody knows everything about the Arduino from the top to the bare metal bottom, let alone something like a car or NASA rocket. So you either wind up being a huge organization or company like NASA or Apple who literally hires lots and lots of people who become experts in their super specialized niche who all work together to achieve a common set of goals or you wind up only knowing macro level items and then basically hoping nothing goes wrong and everything in between. Cost structures, mean time to market and such all differ dramatically here too.

          From the user’s perspective, one of the biggest reasons Arduino and other “non career programmer friendly” types of products have taken off is because it frees you up from having to deal with the underlying “stuff” that bogs a single less technical user down. Which means you can go from idea to working model much more quickly and easily. Sure, it’s not perfect but it at least works. Mostly. Some of the time. But it sure is faster.

      2. I second this sentiment. There are some things I want absolute maximum efficiency for (video encoding in my case) but I don’t have the chops to help out there so let someone else handle that.

        The stuff I write I only care about upper time bound. If it takes an hour to do a task that’s too much, but 5 minutes or as much time as it takes to make a cup of coffee? Good enough! Don’t need to know x86 assembly to hit that level.

        You should still learn it anyway, but that’s the distinction here, you should know about for instance your tax system but shouldn’t have to think about it every time you buy an apple.

    2. Just like how politicians don’t need to comprehend the demographic they say they represent, then do things contrary to common sense and invite the snakes into the country for a party ? because not all snakes are venomous ?!!!.

      1. That’s because politicians are dishonest and serve themselves, but are very effective at lying about it thanks to sophisitcated PR and corrupt media. And also because many voters are fuckin’ stupid.

        Politics is a bad metaphor to use for any example of doing things correctly. Doing them badly and getting away with it, they are ne pas ultra. But politics and human beings don’t work according to logical principles.

    1. See, embedded processors are a huge market. But Intel’s business is selling very expensive chips. So they keep trying to find a place for very expensive embedded processors.

      They’ve no chance. You can get an ARM with a decent fraction of the power of these Intel chips, for a tiny fraction of the price. Fortunately they have the money to keep trying this daft plan, over and over, without going bankrupt.

      I dunno, maybe there’s some PR value in this? I bet Intel spend more on PR and adverts than they have developing these embedded behemoths. Could be.

  3. Looks nifty.

    I think every hacker guy/gal approaching an Intel person showing off things should have the courage to ask uncomfortable questions about embedded subsystem firmware.

    Perhaps, oh perhaps Intel ends up listening.

    Not that most chipmakers are much better, but Intel’s the Godzilla on the block, and at the moment they have some egg on their face (and AMD has been overheard making interesting noises recently). So please, keep at it, and at each occassion, nag them: it’s our duty as hackers!

    1. Yeah! One does not own one’s own hardware, until one owns one’s firmware, and has access to the complete knowlege of the innerworkings of the hardware contained therein.

  4. My question about a ~$325 price tag for the Joule (which seems to be discontinued if I follow the provided link): how does it compare to a Nvidia TX1? After the release of the TX2, the TX1 can be purchased for only $200. While carrier boards seem to be limited, the TX1 has an actual GPU built in, which is essential for getting the most out of applications that need computer vision. While the Joule has 12-18 execution units for graphics, the TX1 has 256 CUDA cores ripe to be utilized for computer vision and machine learning applications.

    If the TX1 could be ordered with a smaller, more minimal carrier board better suited for robotics, it could be a better alternative.

  5. For posterity sake, it’s funny to see both discontinued after less than a year in the market, with hundreds of unsold Euclid’s sitting in stock on various websites, still at full price, but with a big “NO SUPPORT NO WARRANTY” disclaimer.
    Meanwhile you can buy an Intel Realsence board, NEW, online for $25, with a USB 3 breakout for another $25.
    There’s your product, and there’s the profit margin intel’s making on these.
    They could have sold the Euclid for $99-150 and they would have killed the Pi’s robotics marketshare overnight, but no, they needed that 4x profit margin for a glorified Edison strapped to a $25 camera board.
    Thus, for all the talk of “Small margins and low volume production”, and god forbid “SELLING IT AT A LOSS” (Like in the comments above. Remember, that cam module is new for $25 online as a replacement part for laptops, the adapter board to make it useful for hackers costing the same amount), like every maker product Intel’s ever tried before, they were too greedy and ended up loosing everything. They saw makers as the end profit itself, not as a means to get in to a market.
    Maybe they’ll try again some other time. I wonder if they’re gonna make the exact same mistakes they did for the Curie, Quark, Arduino 101, Edison, Joule, Euclid, and make more all over again?

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.