Reinforcement learning guided robotic platform that investigates and discovers asymmetric reactions

Abstract Autonomous robotic platforms accelerate chemical discovery by coupling automated experimentation with algorithmic decision-making, but most lack integrated support for chirality and the sequential decision-making required to explore asymmetric reaction spaces with unknown reactivity and stereochemical outcomes. Here, we report a bespoke robotic platform guided by deep reinforcement learning that autonomously navigates a defined asymmetric reaction space by scouting reaction conditions, discovering chiral products, screening catalysts and mapping substrate scope. The structure-agnostic workflow is enabled by a digital twin of asymmetric catalysis, in which control policies are trained before transfer to the laboratory. We validate the approach in chiral phosphoric acid catalysis: over 30 days, the system made more than 3,000 decisions and uncovered four highly enantioselective transformations, including an axially chiral molecule and three allene syntheses, one involving the enantioselective functionalization of a pyrrolopyrrole moiety. These results highlight deep reinforcement learning-guided autonomy as a strategy for discovery in complex catalytic systems.