Jump to content
[dismiss]
Molecupedia is a parody encyclopedia written by Molecular Reality Corporation. It is not Wikipedia and is not affiliated with the Wikimedia Foundation. About

Self-driving laboratory

Page protected
From Molecupedia, the encyclopedia of very small things
Laboratory in which software chooses experiments and robots run them

A self-driving laboratory (SDL), also called an autonomous laboratory, is a laboratory in which software chooses the experiments, robots carry them out, instruments measure the results, and the results go back to the software to decide what to do next, with little or no person in the loop. A 2024 review in Chemical Reviews puts the promise in two sentences: "Self-driving laboratories (SDLs) promise an accelerated application of the scientific method. Through the automation of experimental workflows, along with autonomous experimental planning, SDLs hold the potential to greatly accelerate research in chemistry and materials discovery."[1]

The idea is older than the current wave of AI companies, which have made it a commercial subject. Lila Sciences and Periodic Labs both build laboratories of this kind, and Molecular Reality Corporation, a nanopore company, calls such laboratories centralized "AI science cathedrals", and counts both companies in its brain lab column.[2]

The loop

A cycle of four boxes: plan, make, measure, learn
The loop of a self-driving laboratory. Software plans an experiment, robots make the sample, instruments measure it, and the result updates the software's model. Schematic; real systems differ in how much of the loop a person performs.

Each turn of the loop has four parts. A planner chooses what to try next, usually with a method that balances exploring new conditions against refining good ones, such as Bayesian optimization or active learning. Robots prepare the sample. Instruments characterize it. The measurement is passed back to the planner, which updates its model and picks the next experiment. What separates a self-driving laboratory from an automated one is the last step: an automated laboratory carries out a list of experiments someone wrote, and a self-driving one writes its own list from the results.

A cloud laboratory, such as Emerald Cloud Lab, is a different thing. It describes itself as "fully software controlled, highly automated life science laboratories", operated remotely through "a single unified software interface", with over 200 instrument models and techniques that are "scriptable and push-button reproducible".[3] A person still decides what to run. Put a planner in the user's place, and a cloud laboratory can become the body of a self-driving one.

Examples

The robot scientist "Adam", from Aberystwyth University, was reported in Science in 2009. It generated hypotheses about gene function in baker's yeast, tested them with laboratory automation, and had its conclusions confirmed by manual experiments. To describe the work the authors built an ontology and a logical language, and the formal record covers over 10,000 research units, nested 10 levels deep, that connect 6.6 million biomass measurements to their logical description.[4]

In 2020 a group at the University of Liverpool used a mobile robot to search for photocatalysts that make hydrogen from water. The robot worked autonomously for eight days, performed 688 experiments in a ten-variable space under a batched Bayesian search, and found mixtures that were six times more active than the starting formulations. Its design, the authors wrote, was "automating the researcher rather than the instruments": a free-roaming robot used the instruments that were already in the laboratory.[5]

The A-Lab at Berkeley Lab, reported in Nature in 2023, synthesizes inorganic powders. Its targets were chosen from phase-stability data from the Materials Project and Google DeepMind, recipes were proposed by language models trained on the literature, and an active-learning loop revised them. In 17 days of continuous operation it made 41 of 58 target compounds, which the authors called novel.[6] A critique in PRX Energy in March 2024 examined the products and concluded that "no new materials have been discovered in that work", adding that "Automated Rietveld analysis of powder x-ray diffraction data is not yet reliable."[7] Rietveld analysis fits a calculated powder diffraction pattern to a measured one, which gives the crystal structures and the amounts of the phases in the powder.

Companies

Lila Sciences describes the instruments of its laboratory as "AI Science Factory" instruments. In a campaign it reported in September 2026, a model proposed catalyst compositions, a deposition system made 96 samples at a time, and the results went back to the model. The laboratory screened 2,942 oxides in three months. For that work "the company opted to use humans to transfer samples from one machine to another".[8]

Periodic Labs says its laboratories in Menlo Park "run 24/7", and lists what limits them: "The number of agents is not elastically scalable", "each experiment can take days", and "the results are often ambiguous".[9] The company has also trained a model, Periodic Neon, to interpret X-ray diffraction results, the same kind of data whose automated analysis the A-Lab critique questioned, and reports a 55.3 percent success rate on its most difficult internal evaluation set.[10]

In Molecular Reality's usage

Main article: Brain lab

Molecular Reality's materials propose small sensors in homes, sinks and drains against the "AI science cathedrals".[2] Its slogan for the alternative is "Ten billion wet labs."[2] The proposal, utility-scale molecular sensing, is a network of household molecular sensors. In August 2026 the company's field guide gave the present size of the effort in one sentence: "Today we have one instrument in one lab."[11]

See also

References

  1. ^ Tom et al. (2024). "Self-Driving Laboratories for Chemistry and Materials Science". Chemical Reviews 124: 9633-9732. doi:10.1021/acs.chemrev.4c00055.
  2. ^ a b c "Company records". Molecular Reality Corporation. October 2026.
  3. ^ "Emerald Cloud Lab". Emerald Cloud Lab. Retrieved October 25, 2026.
  4. ^ King et al. (2009). "The automation of science". Science 324: 85-89. doi:10.1126/science.1165620.
  5. ^ Burger et al. (2020). "A mobile robotic chemist". Nature 583: 237-241. doi:10.1038/s41586-020-2442-2.
  6. ^ Szymanski, Nathan J. et al. (2023). "An autonomous laboratory for the accelerated synthesis of inorganic materials". Nature 624: 86-91. doi:10.1038/s41586-023-06734-w.
  7. ^ Leeman et al. (2024). "Challenges in High-Throughput Inorganic Materials Prediction and Autonomous Synthesis". PRX Energy 3: 011002. doi:10.1103/PRXEnergy.3.011002.
  8. ^ "How an AI-run lab cracked open green hydrogen's catalyst problem". Lila Sciences. September 25, 2026.
  9. ^ "Building Labs that Learn". Periodic Labs. September 15, 2026.
  10. ^ "Nature Is Our Learning Environment". Periodic Labs. September 15, 2026.
  11. ^ "Keltar's Field Guide to Molecular Reality". Molecular Reality Corporation. August 2026.
Retrieved from "https://molecularreality.com/wiki/Self-driving_laboratory"
Last edited 1 hour ago by PicoampPedant