A Polish Academy of Sciences lab robot identified four previously undescribed enantioselective reactions during a 30-day run, according to work published in Nature Communications on October 13, 2026. Researchers chose the starting materials and defined the search space, then isolated and characterized the products by hand to confirm the findings.
The robot used reinforcement learning to choose what to try next, including whether to refine a promising result or move on. It made more than 3,000 decisions, which the authors caution should not be counted as 3,000 full experiments. The team trained the strategy in a simulated “digital twin” before using it in the lab to limit reagent use and instrument time.
