Nature Machine Intelligence, Published online: 28 September 2026; doi:10.1038/s42256-026-01305-w

A vectorized simulator and structured reward framework enable microrobot navigation policies to be trained within minutes and transferred without retraining across robots and environments.