Self-Driving Labs
technology Also: autonomous laboratories, lab automation, autonomous experimentation, closed-loop experimentation.
A self-driving lab closes the loop of experimental science: a model proposes an experiment, robots and instruments carry it out, the measurement goes back to the model, and the model picks the next experiment.
What it is
The aim is many more experiments per researcher, and searches through chemistry or biology too large to cover by hand.
Where it sits
Where it stands (Nov 2026)
- Biology joins materials. A November 2026 Nature paper describes a biology lab where AI plans and runs work across more than 20 kinds of instrument and over 11,000 steps, from making DNA to sequencing it, with humans only restocking consumables; it matched expert yields and ran protocols 3.2 times faster.
- The lab that lays itself out. “Self-Building Labs” (Aspuru-Guzik group, Toronto, October 2026 preprint, not peer reviewed) uses AI agents to turn a research description into a robot protocol, test it in a simulated copy of the lab, then drive the real arm; in one test it wrote more than 200 commands and ran a real nanocrystal measurement.
- Orchestration software is the active layer: schedulers that share instruments between runs and agent operating systems for materials research.,
What decides the pace
The cost of instruments that accept robot handling, the step from one tuned workflow to a lab that runs many, data formats instruments share, and whether results from automated runs reproduce in other labs.
What would change the view
- Is the bottleneck the robot, the protocol writing, or the instruments’ interfaces?
- Who sells it: instrument makers, cloud-lab operators, or software layers on top?
- Does biology, with messier samples, follow the materials curve or stall?