For the last six months I've been tracking who is building self-driving labs, the AI-driven, robotically operated science platforms that promise to close the loop between hypothesis and experiment. The result is a hand-curated interactive map at sdl-map.discoverylabs.nl. 74 entries across four tiers: national programmes, academic groups, commercial vendors, and lab-OS platforms.
This is not a search-indexed listing. It captures the landscape as I read it. Which countries are putting public money behind autonomous discovery (Canada's Acceleration Consortium, Korea's MOTIE 500-lab plan, France's PEPR DIADEM, the EU's BIG-MAP / FULL-MAP). Which academic groups are actually running platforms (A-Lab, Cooper, Cronin, Jensen, Abolhasani, Jun Jiang). Which companies moved past pitch decks into operating facilities (Lila Sciences, Medra, XtalPi, Cusp AI). And which OS layers quietly orchestrate everything underneath (Automata LINQ, Benchling, Artificial, Atinary).
I built it because the strategic question "where is this field clustering, and where are the gaps?" was not answerable from any single source. Funding announcements live in press releases; group capabilities live in their websites; vendor footprints live in industry reports. The map fuses them. The repository is open at github.com/yetiswang/sdl-map. Every entry's metadata, location, funding label, and sources are in the JSON, free to fork.
What I see when I step back: characterisation-in-loop is still the rarest capability. Most operating SDLs do synthesis well and characterise crudely. Only a handful integrate advanced multimodal characterisation into the autonomous loop. That's the gap worth watching.