Materials scientist here, your post may be underestimating how difficult it is to make competitive scientific hardware. It is a lot easier for me to see Bruker adding a python API to their x-ray diffractometers (XRDs) than for another company to make an "automation-native" XRD. Goes double for more complex microscopy / optics, maaaybe easier for simpler measurements. I think this should generalize: wherever the existing hardware is relatively high-tech in itself, the incumbents should have an advantage over newcomers in their ability to just add an API.
Essentially autonomy-native hardware harnesses for cheap existing characterization equipment + better software API's on top for integration. But would definitely love to chat more about it!
And while the trend you note is generally true, the entire "software is eating the world" thesis is that incumbents will eventually be disrupted (or forced to innovate)
I think I see what you mean. There is indeed a current of research in academia to make routine measurements cheaper / faster via open-source-like platforms. For example, in electrochemistry, Joaquín Rodríguez-López at UIUC and of course Alán at Toronto come to mind. We will see how that unfolds, how much sticks, and how much becomes commercial.
Oh I hadn’t followed Joaquín Rodríguez-López’s work, will check out.
Yeah I think open question to what extent this is relegated to the fringe of open source science tinkering vs how to build commercial trust in these kinds of solutions that enable high data quantity over quality, especially given so much of the industry is just downstream of biotech
I'm not sure what "downstream" means in this context. A lot of materials science (I've worked on solar water splitting, PV, and batteries, batteries the more difficult) has more diverse / complex synthesis protocols, and more complex characterization, with frontier science oftentimes requiring multimodal characterization, simply by virtue of dealing with mostly solids rather than mostly fluids.
The DoE solar hydrogen (afterwards CO2) hub had a high-throughput experimentation team from its 2010 start, and e.g. John Gregoire went from that field to Lila recently.
Materials scientist here, your post may be underestimating how difficult it is to make competitive scientific hardware. It is a lot easier for me to see Bruker adding a python API to their x-ray diffractometers (XRDs) than for another company to make an "automation-native" XRD. Goes double for more complex microscopy / optics, maaaybe easier for simpler measurements. I think this should generalize: wherever the existing hardware is relatively high-tech in itself, the incumbents should have an advantage over newcomers in their ability to just add an API.
100% agree, the hardware moat is quite high. What I was thinking about was something like this (https://pubs.rsc.org/en/content/articlelanding/2024/mh/d4mh00797b) or this (https://ojs.lib.uwo.ca/index.php/openhardware/article/view/23044)
Essentially autonomy-native hardware harnesses for cheap existing characterization equipment + better software API's on top for integration. But would definitely love to chat more about it!
And while the trend you note is generally true, the entire "software is eating the world" thesis is that incumbents will eventually be disrupted (or forced to innovate)
I think I see what you mean. There is indeed a current of research in academia to make routine measurements cheaper / faster via open-source-like platforms. For example, in electrochemistry, Joaquín Rodríguez-López at UIUC and of course Alán at Toronto come to mind. We will see how that unfolds, how much sticks, and how much becomes commercial.
Oh I hadn’t followed Joaquín Rodríguez-López’s work, will check out.
Yeah I think open question to what extent this is relegated to the fringe of open source science tinkering vs how to build commercial trust in these kinds of solutions that enable high data quantity over quality, especially given so much of the industry is just downstream of biotech
I'm not sure what "downstream" means in this context. A lot of materials science (I've worked on solar water splitting, PV, and batteries, batteries the more difficult) has more diverse / complex synthesis protocols, and more complex characterization, with frontier science oftentimes requiring multimodal characterization, simply by virtue of dealing with mostly solids rather than mostly fluids.
We have seen a lot of high-throughput science as folks searched for catalysts and photocatalysts, for example this work predates any SDLs I know of https://pubs.rsc.org/en/content/articlelanding/2009/ee/b812177j
The DoE solar hydrogen (afterwards CO2) hub had a high-throughput experimentation team from its 2010 start, and e.g. John Gregoire went from that field to Lila recently.
Great article. Do you have similar information (list) about self-driving labs based in Universities globally?
I have the U.S., Canadian, and U.K. relatively well mapped out, but a global one would be helpful.
This is a helpful review of self-driving labs in China:https://pubs.rsc.org/en/content/articlelanding/2025/dd/d5dd00072f
It would be tremendously helpful to have some visualization and analysis of funding/development of self-driving labs across the globe.