Fold Commons

MoleculeFlow

How a self-driving lab actually runs a discovery campaign — as a loop, not a straight line. Each landmark run below is broken into the same five stages an autonomous system cycles through (plan → synthesize → measure → analyze → decide), and every one is cited to its primary paper. Nothing is installed; nothing about you is collected.

This is the tool — a deliberately small curated starter set of published runs, not a comprehensive index — see the coverage gauge below. Editorial is CC BY 4.0; each run links its primary paper by DOI. A native version, when it ships, adds offline use; it never gates the web.

The closed loop

1 Plan 2 Synthesize 3 Measure 4 Analyze 5 Decide
  1. 1Plan — Propose the next experiment
  2. 2Synthesize — Run it on the robotic platform
  3. 3Measure — Read the instruments
  4. 4Analyze — Interpret the result
  5. 5Decide — Accept, reject, or iterate

A curated starter set — not a comprehensive index

This edition tracks 3 landmark runs from the published autonomous / self-driving-lab campaigns. Nobody counts every such campaign, so this is a count, not a share of the field. We show what we curate and name what we omit, so you can calibrate. As of 2026-06-10.

What’s covered

  • Landmark autonomous / self-driving-lab campaigns published in peer-reviewed venues or arXiv
  • Closed-loop runs with a legible plan → synthesize → measure → analyze → decide cycle
  • Runs whose primary paper is openly citable by DOI

Known gaps

  • Most published self-driving-lab campaigns (ChemOS / Gryffin / Phoenics, Ada, RoboRXN, Emerald Cloud Lab, Berkeley A-Lab follow-ups) not yet curated here
  • Materials-acceleration-platform (MAP) runs beyond the single A-Lab example
  • Industrial closed-loop pipelines that publish no methods paper
  • Negative or aborted campaigns (survivorship bias toward headline successes)
  • Per-step raw instrument data — this corpus curates the loop structure, not the underlying datasets
Lab
Year

3 of 3 runs

  1. A-Lab 2023

    A-Lab makes inorganic materials on its own

    An autonomous synthesis platform made inorganic materials over 17 days of largely unattended operation. The first report described many of them as new compounds; that claim and the automated phase identification were challenged in 2024 (Leeman et al., PRX Energy), and Nature published an author correction to the paper in January 2026.

    1. 1. Plan DFT-screen candidate stoichiometries; rank by predicted stability.
    2. 2. Synthesize Robot weighs precursors, mixes, and fires them in a tube furnace.
    3. 3. Measure Run XRD on the quenched powder; cluster phases against the ICSD.
    4. 4. Analyze Refine cell parameters; flag phases that match no known entry.
    5. 5. Decide Promote the phase to the catalogue; queue follow-up characterisation.

    Primary paper Szymanski et al., Nature (2023) doi:10.1038/s41586-023-06734-w

    Also on the record

  2. Coscientist 2023

    Autonomous Suzuki–Miyaura coupling planning

    A GPT-4-driven planner ("Coscientist") designed and ran Pd-catalysed Suzuki–Miyaura and Sonogashira cross-couplings on a liquid-handling robot. The authors note the setup was not yet fully automated (plates were moved by hand), but no human made a decision.

    1. 1. Plan Propose a Suzuki–Miyaura coupling between an aryl halide and a boronic acid.
    2. 2. Synthesize Write liquid-handler instructions; run the reactions in a plate on a heater–shaker.
    3. 3. Measure Analyse the reaction mixtures by gas chromatography–mass spectrometry (GC–MS).
    4. 4. Analyze Look for the target product's signal in the chromatogram and its mass spectrum.
    5. 5. Decide Accept the route; archive the successful protocol.

    Primary paper Boiko et al., Nature (2023) doi:10.1038/s41586-023-06792-0

  3. ORGANA 2024

    ORGANA runs heterogeneous chemistry with a modular agent

    A composable planning + execution graph drives wet chemistry across multiple instrument backends without per-task retooling. First posted on arXiv (2024); the peer-reviewed version is in Matter 8:101897 (2025).

    1. 1. Plan Compose a plan over the reagent + instrument graph.
    2. 2. Synthesize Route subtasks to the liquid handler / glovebox / electrochemistry rig.
    3. 3. Measure Stream sensor readings into a central state store.
    4. 4. Analyze Apply a per-task analysis tool; emit a structured outcome.
    5. 5. Decide Update the plan; loop until the success criterion is met.

    Primary paper Darvish et al., Matter (2025) doi:10.1016/j.matt.2024.10.015

    Also on the record