A self-driving lab loop: a gantry liquid handler dispenses reaction mixtures onto a plate, a coiled flow-chemistry reactor carries each one through continuous flow, and an in-line analyzer reads back the yield. A Bayesian optimizer proposes the next concentration/temperature pair from a live response-surface model (mean estimate plus an exploration bonus), converging on the true optimum in far fewer runs than random or full grid search.
mean + kappa·sqrt(variance) — high where the model expects a good yield, or is still very uncertain. A few opening samples are always random to seed the model.Real self-driving labs already do this at scale: UC Berkeley's "A-Lab" combined robotics with machine-learning proposals to attempt 355 novel inorganic materials and successfully synthesize 41 of them in just 17 days of autonomous operation, a pace that would take a human team years of manual trial and error.
A self-driving lab loop: a gantry liquid handler dispenses reaction mixtures onto a plate, a coiled flow-chemistry reactor carries each one through continuous flow, and an in-line analyzer reads back the yield. A Bayesian optimizer proposes the next concentration/temperature pair from a live response-surface model, converging on the true optimum in far fewer runs than random or full grid search.
A closed-loop autonomous experimentation cycle: dispense, flow, analyze, and feed the result back into a Gaussian-process surrogate model. The Bayesian optimizer balances exploiting conditions the model already expects to be good against exploring conditions it is still uncertain about, which is why it reliably beats random search to the true reaction optimum.
Pick Bayesian or Random search, set the iteration budget, then click Start optimization run and watch the gantry dispense, the flow reactor carry the mixture, and the analyzer read the yield each cycle. Run both strategies back to back to compare their convergence curves on the shared chart, and use Reset to clear everything and start over.
Real self-driving labs already do this at scale: UC Berkeley's "A-Lab" combined robotics with machine-learning proposals to attempt 355 novel inorganic materials and successfully synthesize 41 of them in just 17 days of autonomous operation.