The Raspberry Pi AI kit is arguably one of the most sensible consumer-grade AI products released yet. That isn't because it's the fastest AI accelerator—far, far from it—but because it targets two critical issues facing the world of artificial intelligence technology.
Unlike most AI hardware, it's affordable ($70, not counting a Raspberry Pi 5 that you'll need to mount it on), which makes it far more accessible than most of the gear out there. And since it's for the Raspberry Pi, a device all about helping people learn to program, the kit is an excellent platform for getting coders ready for AI. Given its specialized nature, we didn't have a comparable or competitive product to review against it, but we can definitely say that the Raspberry Pi AI Kit has a lot of promise.
Unboxing the Raspberry Pi AI Kit
The $70 kit includes a Raspberry Pi M.2 HAT+ (an auxiliary board you'll connect to your Pi 5), a Hailo AI accelerator with a Hailo-8L AI chip, and the necessary mounting hardware to connect the HAT+ to your Raspberry Pi 5 (which you'll need to purchase separately for $80). You can buy the components separately, but you'll almost certainly pay more than $70 for the Hailo-8L AI accelerator alone.
(Credit: Michael Justin Allen Sexton)With a reported 13 TOPS of AI processing power, the Hailo-8L stacks up well against other first-generation AI accelerators like the ones inside AMD's Ryzen 7040 series mobile processors and Intel's "Meteor Lake" laptop chips (which are rated for 10 and 11 TOPS, respectively). Announced next-generation components like AMD's upcoming Ryzen AI 300 and Intel's Core Ultra "Lunar Lake" mobile chips will have a fair bit more, but 13 TOPS is enough to be getting on with for now.
(Credit: Michael Justin Allen Sexton)Putting the components together is easy and straightforward. You just need to attach the GPIO header, place the HAT+ on the Raspberry Pi 5, and then screw in the mounting hardware (which consists of four standoffs with screws) in the corners.
(Credit: Michael Justin Allen Sexton)This device doesn't use the GPIO pins for any purpose; the header is simply there to make connecting other devices to the GPIO easier with the HAT+ installed. Instead, the HAT+ connects using the Raspberry Pi 5's single PCI Express 2.0 lane.
The M.2-based AI accelerator comes preinstalled on the HAT+, so all that's left to do is connect the ribbon cable. This, too, is preinstalled on the HAT+, but I found it easier to disconnect it there and connect it to the Raspberry Pi 5 first. The way the cable fits into the connectors on the two devices makes this a bit easier.
(Credit: Michael Justin Allen Sexton)Technically, that's it for the hardware setup. From there, you can install the software and begin using the device, which we'll discuss more in the next section. If you want to try out the demo software the Raspberry Pi Foundation provides, you'll also need to connect a camera, such as one of Raspberry Pi's camera modules. (We used the Raspberry Pi Camera Module 3 pictured earlier.) This can be attached after the HAT+ is installed; that board even has a cutout for the cable to pass through it so there are no issues with adding a camera later or removing one if you no longer plan to use it.
(Credit: Michael Justin Allen Sexton)No other hardware is strictly necessary, but if you want to equip your Raspberry Pi with a heatsink or fan, the installation leaves you room to do so.
Software Setup & Exploring Demos
With everything assembled, you're ready to tackle the software to make use of your AI hardware. The company recommends you begin—before installing the Pi AI Kit—by ensuring your Pi 5 is running the latest version of the Raspberry Pi operating system, along with the latest updates. We certainly agree with this advice and followed it before attaching the HAT+.
Next, you need to install the drivers and software needed for the Hailo AI accelerator to operate. You accomplish this by typing this command into a terminal...
$ sudo apt install hailo-all
...and rebooting the device after the installation. If you want to try out the included demos, you'll need to acquire those using the following command to download them...
$ git clone - -depth 1 https://github.com/raspberrypi/rpicam-apps.git ~/rpicam-apps
Currently, three demos are available that focus on AI usage with the camera. Raspberry Pi calls the first of these a pose estimation demo. It feels readily familiar to anyone who's owned an Xbox Kinect or similar device, as it essentially tracks your body movements by following 17 points across your body. The demo worked just fine when I tested it, tracking my arm and body movements with reasonable accuracy.
Second is an object detection demo that attempts to identify items that you hold in front of the camera. This, too, seemed to work reasonably well, but with more errors, sometimes giving a wrong answer first or switching answers. It knew, for instance, that I was a person and that my cat was a cat, but thought one of my screwdrivers (with an admittedly small handle) was a pen.
The third demo is focused on image segmentation and allows you to isolate objects, the idea being that afterward you can strip out the object or background and replace it with something else.
Verdict: Still Under Development
All three of the supplied demos worked with only minor mistakes that weren't too far off of the mark. As is, the Raspberry Pi AI kit is fun to play with, but there's little reason to buy one unless you plan to use it to learn to create programs or devices that take advantage of the AI power.
(Credit: Michael Justin Allen Sexton)At its core, the Raspberry Pi 5 and new AI kit are meant for use by students and hobbyists who want to learn to program and create their own semi-custom devices, and we see no reason why the kit shouldn't be fine for that purpose. It doesn't have much to offer (at least for the time being) if you're not looking to develop your own code, but it's an excellent way to get the ball rolling on independent AI software development with a low cost of entry.