It’s on the tip of every clueless exec’s tongue lately: AI! And while the nauseating levels of over-hype may have caused you to lose your appetite for the topic completely, there are actually some interesting, legitimate use cases for artificial intelligence, especially in the embedded space. So this week we’re taking a look at some new AI-focused hardware that actually does something of value rather than just hallucinating some Cronenbergian “art” at you while plundering our natural resources.

Table of Contents

• Pi AI

• BBY-AI

• A1101

• Teardown

• The Zen of Maker Faire Kyoto

Pi AI

Adding brains to a Raspberry Pi via a new HAT (gosh, wouldn’t that be great for humans?!) is not unprecedented, with Adafruit, Google, Pineboards, Seeed and more having produced various machine learning augmentation and acceleration solutions in recent years. But when Raspberry Pi produces their own specialty board, it’s a pretty good indication that the topic to which it pertains is worthy of folks’ attention. As such, we were incredibly excited to get our hands on the new Raspberry Pi AI Kit, which is an M.2 HAT+ like you may have seen for adding NVMe drives to the Pi 5, but in this case with a 2242-sized AI module jammed in the slot. The accelerator module is built around the Hailo-8L neural processing unit (NPU), which is then automatically made available to AI tasks and the built-in camera applications, thanks to the latest Raspberry Pi OS, adding a whopping 13 tera-operations per second (TOPS) of AI-optimized compute (more than three times Google’s Coral USB Accelerator, for example). In short: big AI power with Raspberry Pi ease, available now at Digi-Key.

BBY-AI

Elsewhere in credit-card-sized computing, our good friends at BeagleBoard have released their artificial intelligence-focused BeagleY-AI board. You may have caught the board’s debut at Embedded World a few months back as part of Anouk Wipprecht’s DRINKBOT 2.0, but now you can get your hands on your own with the official release this month. Unlike the earlier, yet more powerful BeagleBone AI-64, the BeagleY-AI eschews BeagleBone cape compatibility and takes the familiar form factor of the modern 40-pin Raspberry Pi. While its 4 TOPS of AI acceleration may not sound like a ton next to the Pi AI Kit’s 13, the BeagleY board gives you a complete solution for around $70, compared to the AI Kit, which is $70 alone, and requires a $60-80 Raspberry Pi 5 to operate, making it roughly twice the price overall. And since the BBY-AI is Pi-HAT-compatible, you could even stick an AI Kit on it, so you could use the built-in C7, plus the Hailo-8l, and hey, why not shove a Coral in the USB too, for a theoretical 21 TOPS (sounds far-fetched, but I talked to a Ti engineer who said it should be feasible!). Start your journey from “why AI?” to BeagleY-AI at digikey.com!

A1101

One interesting development I’ve noticed lately is the proliferation of application-specific AI hardware, rather than the usual “here’s some AI in a box, go make something cool.” We covered Seeed’s Grove Vision AI v2 Kit vision module earlier this year, with its zero-code Sensecraft AI capabilities, and lately we’ve been playing with another highly-specialized device: the SenseCAP A1101 LoRaWAN Vision AI Sensor. The A1101 pairs a HiMax ultra-low-power image sensor with a LoRa radio and waterproof enclosure, meaning you can set it up, then go stick it in a field or jungle for half a year, and not have to worry about charging or power. In addition to great tutorials from Seeed, the A1101 also has full support from Edge Impulse, which makes machine vision tasks like object detection an absolute breeze. Pick one up for yourself from Digi-Key today!

Teardown

If this newsletter feels rushed (hopefully not!) it’s because I spent the past weekend and then some at Crowd Supply’s annual Teardown event in Portland, Oregon.

Held in a former Gap store in a semi-abandoned mall, Teardown was three glorious days of open-source hardware hacking, learning, and sharing, among around 300 rad artists/engineers/designers/educators/enthusiasts. I learned how to use Krita to enhance my KiCad PCB designs, how to build a Jumperless Probe and then poke stuff, how to make my badge life easier with an open-source template, and at long last, how to program my Supercon 2022 Voja4 badge – plus so much more! If you’d like to get a taste of what Teardown has to offer, you can watch a livestream I did with organizer Helen Leigh:

or check out the official web site. I would imagine the team are going to be recovering for a bit, but eventually the talks should be available to watch on their YouTube channel, where you can also find a bunch of great stuff from last year. And if any/all of this appeals to you, I really hope to see you at Teardown 2025!

The Zen of Maker Faire Kyoto

The team from Maker Faire Kyoto just published their video from their April 2024 event.

That’s it for this month! I’d love to hear about your machine learning or other electronics projects – drop me at note at [email protected] and tell me what you’re up to!

Until next time,
David J. Groom, person in desperate need of sleep after an amazing Teardown!

Make Things is a weekly newsletter for the Maker community from Make:. This newsletter lives on the web at makethings.make.co