…Looking To Develop AI/ML Applications On The EDGE?
Arthurtm is a Raspberry-Pi Hat for Infrared based AI & Machine Learning development. It’s designed to work with any programming language able to address the RPi-5/4B/3B+ & 3B GPIO pins and Camera Serial Interface, whether natively or through the OS. It isn’t locked-in to any specific development framework. The Raspberry Pi-5/4B/3B+ can run all the usual AI/ML libraries like OpenCV, TensorFlow, PyTorch, TF/Keras, including AI/ML stacks that don’t require GPU or TPU hardware.
Can I use Arthurtm with TensorFlow?
If your preferred framework can run on ARM, then you’re in business. While the RPi’s lack of available GPU/TPU implies you likely won’t get real-time performance, there’s a surprisingly large class of practical applications that are well within the capability of the RPi-5,4B & 3B+, at an affordable price point, and with minimal fuss.
What’s the Angle?
But the main motivation for designing (yet) another RPi Hat stems from a different need. We wanted a Hat with the right specs (see below) that can run AI/ML models trained on the RPi itself. We believe that Edge applications have a very bright future. Specifically, TensorFlow & TFLite models trained in the Wolfram Languagetm API. This framework is based on the supervised & unsupervised Machine Learning “super-functions” Classify & Predict and FindClusters & FeatureExtraction, respectively.
Why Wolfram Language?
Because you can easily create your own models using surprisingly modest data sets, with pleasantly effective performance. And you can train them directly on the Raspberry-Pi. The development cycle is extremely rapid (thanks to Wolfram’s unified computational language framework) with good, practical results. – No need to download and adapt large AI/ML models from the “zoo” that require training on sophisticated hardware, or having to go through the steep learning curve of mastering cross-platform / cross language model adaptation.
Wolfram Mathematica (or Wolfram Alpha)
Thanks to Wolfram-Research, both the Wolfram Language engine, (including Mathematicatm) & license, come bundled free (for personal use) with Raspbian OS. You can even interface Arthurtm with Wolfram Alpha’s API. The Raspberry-Pi OS is available for download at the Raspberry-Pi foundation website.
Julia, Python, (insert your favorite language here…)
We are also doing some development work with the Arthurtm Infrared AI/ML Hat and Julia Language AI/ML frameworks like Flux.jl, Knet.jl, MLJ.jl and Metalhead, ObjectDetector etc. The idea being to leverage Julia’s unique combination of software development speed and high runtime performance for model training and deployment. All without leaving the comfort of the Raspberry-Pi development environment. You can even use Python.
Arthurtm Infrared AI/ML Dev Hat:
- Is fully RPI-HAT compliant. Onboard EEPROM assigns audio output from RPi audio jack to the Hat through the 40-pin GPIO header.
- Can mount any RPi NoIR or color CSI camera on the front side of the Hat.
- Powers the RPi-5/4B/3B+ etc through the 40-pin GPIO header, via the DC barrel jack.
- Comes supplied with an OBD2 power cable for automotive applications.
- Has a Texas Instruments, high-efficiency, dual-phase buck-converter that can source a maximum of 6A at a nominal 5.0 volts. More than enough power for all on-board systems with a good margin for future expansion. Requires 12 volt, 30 watt input at full load. For desk-top development we suggest you buy a suitable 12VDC, 30W wall-wart (not included).
- Includes a general purpose 3V linear regulator output on its own 3 pin header.
- Breaks out all UART, SPI0, SPI1/I2S & I2C ports on separate pin headers.
- Has a separate 1×3 pin, diode & transistor buffered, servomotor port driven by the PIGPIO library/daemon (excellent performance with no jitter).
- Has separate 2xCMOS switch ports ~ 4ohm impedance.
- Has an on-board 3W, Texas-Instruments monolithic linear stereo audio amplifier. Just add speakers.
- Has 2xHigh-power SMD, 850nm wavelength IR-LEDs for object/face illumination.
- Has 8 diode & transistor buffered, digital output lines, including 4 with high-quality Red, Green, Amber & Blue status LEDs.
- Has 3xGeneral purpose, fully assignable, momentary SPST switches.
- Audio input can be accommodated via standard USB sound-card/USB microphone (not included).
- Software based light metering through the CSI camera.
- IMPORTANT: Do Not Apply Power to the Raspberry-Pi (through its micro-USB or USB-C power socket) AND through the Arthur IR AI/ML Hat simultaneously. You may damage your devices.
The Arthurtm Infrared AI/ML Dev Hat comes with complete source-code which includes the following Wolfram Languagetm routines:
- A CLI application to acquire and train a word recognition model based on a list of spoken words (the size of the list is limited only by the available system RAM). We recommend a 4 or 8 GB RPi-5 for best results or RPI-4B for good results.
- A demo application (Wolfram Mathematicatm notebook) for multi-factor authentication (based on PIN, access word, and face recognition) for secure access to your car through it’s central locking system. Includes a comprehensive user guide. With care to detail you can train your model to positively ID any feature (including faces) under a broad range of natural or artificial light conditions including at dusk, dawn, in bright daylight or zero-light. Perfect for serious developers.
- The base demo incorporates encoded, IR illuminated human face data, to train a classifier able to detect and recognize YOUR (or anyone’s) face in low-light or no-light conditions.
- Train, mix and match audio & visual classifiers in the same application. Limited only by the available system RAM. Again we recommend a 4 or 8GB RPi-5/4B.
- More advanced developers can modify the base demo application to train a custom model that can detect and recognize almost any feature using the many thousands of powerful AI, ML, DSP and other advanced functions, commands & routines built right into the Wolfram Languagetm engine.
- With some experience using the Arthurtm v2 Hat Dev board with Wolfram Language, you will be able to customize or write your own source code to solve many practical AI/ML applications. From academia and R&D, to industrial automation, health, education, automotive, transportation, aerospace and more.
Availability – Now Open for Pre-Order
At the time of this writing we only have a very limited supply of the Arthurtm v2 beta Hat boards for sale, due to the on-going chip shortage and supply chain issues beyond our control. However supply should start to improve by Q1, 2024. Production units will be electrically the same with perhaps only some minor or cosmetic changes.
We expect the retail price of the Arthurtm v2 Infrared AI/ML Dev Hat & Software bundle to hover around US$389.99, depending on daily electronic component spot price fluctuations including foreign exchange rates, commodity prices etc. As shortages begin to ease we expect to achieve sufficient inventory levels and prices will likely stabilize (at least that’s what we hope).
Your package will include the assembled Hat board, an OBD2 cable to male DC barrel plug, a code # with a link where you can register and download Arthur v2 source code / Dev tools and user guide PDF.
Register your pre-order for the Arthurtm v2 Hat through the pre-order form below.
Also, briefly add your desired order quantity and your geographic location. Thank you.