A self-driving laboratory, also called an autonomous lab or self-driving lab, is a research setup that closes the loop between deciding what experiment to run and actually running it. An AI system proposes the next experiment, robotic equipment carries it out, instruments measure the result, and a learning algorithm uses that result to choose the next experiment, repeating with little or no human in between.
The concept fuses several threads: active learning, which picks the most informative experiment to do next; laboratory automation and robotics, which handle the physical work; and increasingly large language model agents, which can read the literature, write control code, and reason about results. The aim is to attack the slowest part of experimental science, the trial-and-error cycle, by running it continuously and intelligently.
Concrete demonstrations arrived in force around 2023. Berkeley’s A-Lab synthesized dozens of new inorganic compounds over 17 days, choosing recipes from computed databases and refining them with active learning. Coscientist, built on GPT-4, planned and ran chemistry experiments on robotic hardware. Both showed the closed loop working end to end.
For a business or general reader, the self-driving lab is the organizing vision behind much of modern AI for science: not a single clever model, but an automated discovery engine whose throughput is limited by compute and robotics rather than by human hours. It also brings new questions about reproducibility, verification, and safe use.