SDL-Demo: A teaching kit for autonomous research in the physical sciences

Hi all,

Curious to hear some feedback from the GOSH community about an open-source hardware+software kit that I’m working on: SDL-Demo. In my field, “self-driving” (i.e. autonomous) labs dynamically respond to inputs, similar to a self-driving vehicle, in a closed-loop fashion. In chemistry and materials science laboratories, this involves automated synthesis, automated characterization, and artificial intelligence to accelerate materials discovery through closed-loop experimentation (source). In a response to a call for papers (still open btw), I proposed a minimal working example for self-driving labs, a sort of “Hello, World!” introduction. In many ways, the demo is a combination of “Hello, World!”-s from electronics (blinking a light, reading data), and the optimization community (grid search vs. random search vs. Bayesian optimization). At its core, the demo is a color matching demo using an RGB LED, a spectrophotometer, and adaptive design algorithms to iteratively converge on the best color match. Both the hardware and the code are modular and extensible, making it a nice candidate for teaching and prototyping.

I made the demo remotely accessible via the cloud, meaning you can blink an LED in my home in Salt Lake City, read the light sensor data, and wrap it in a closed-loop optimization scheme via this Google Colab notebook. I made it so multiple people can request experiments simultaneously. I’m also working with CrowdSupply to explore the possibility of hosting this as an assembled kit with pre-loaded code.

Project Images

Project Links

What are your thoughts?


Interesting… allthough for some parts of the world we should try to reduce the costs to below 10$…

How is your experience with that multi-channel spectral color sensor?

The “spectrometer” challenge is a classic educational format for open hardware and science / lab equipment. i started a “module under dev” topic once here in the forum:

also described in this article.

1 Like

@dusjagr thanks for the reply!

A lot of the design considerations that put it at ~$45 have to do with modularity, extensibility, and aesthetics. While it might not align as well with a “Hello, World!” for self-driving laboratories, I think there’s certainly a use-case for and a way to reduce the cost below $10 while keeping the setup time low. For example, using a two-wire LED and a two-wire photodiode (brightness matching instead of color matching) with alligator clips or a mini breadboard instead of Grove/Stemma-QT connections, and using a Pico H instead of a Pico W. See the snippet from the white paper below:

In the simplest setup, a single LED with a single brightness sensor could be used; however, this is missing qualitative features of SDLs for real-world tasks involving multiple tunable inputs and multiple signal measurements; it also presents additional hardware challenges and design considerations. For example, we wanted to keep the signal (i.e. LED) and sensor on separate boards while attached via a cable rather than integrating everything onto a single printed circuit board because it better mimics the SDL best practice of modularity [8, 11]. …

What do you think might be an optimal bill of materials for a separate version with a design constraint of < $10?

Great! The sensor seems fairly reliable and the physical Stemma-QT connection and the CircuitPython library are easy to use. The datasheet had great info, and Adafruit has nice tutorials for it. I ended up needing to use MicroPython for the project, and there’s a nice MicroPython alternative for the AS7341 library on gitlab that someone put together. There’s also an Arduino library.

Thanks for pointing this out! I appreciated the “New dangers and limits” section, which was a topic that panelist Michelle Murphy brought up at the Accelerate conference in one of the panel discussions IIRC. Seems like a very nice workshop with some great learning outcomes.