A University of Toronto-led team developed a system of AI agents that turns scientific goals into simulated lab plans, prepares the skills and workflows needed to run them, and directs physical robots through experiments, including a demonstration that made quantum dots.
The system is called Self-Building Labs, or SBL. It aims to address a weakness of many automated labs. These labs can run only procedures and use equipment they were specifically set up to handle.
The agents identify what a new task requires. They create a digital model of the workspace. They develop ways for robots and instruments to carry out the work.
Before physical experiments, the system can test operations in a simulation. During experiments, safety checks and sensory feedback help guide the work. The researchers report testing SBL with different robot configurations.
These ranged from a single workstation to setups where several robots work together. In the quantum-dot demonstration, the system developed steps for handling liquids, moving and sealing laboratory containers, and taking measurements using light. People still set the scientific goal and approve plans.
This means the system is not a fully independent scientist. The findings are described in an arXiv preprint. The preprint is an early-stage report.
It has not yet established how reliably the approach will work in other labs and experiments.