AI system schedules observations on Blanco Telescope

Scientists at University of Chicago, Fermilab and Northwestern University have developed an AI tool that automatically decides where a telescope should point, then adjusts the plan as weather, moonlight and atmospheric conditions change. The system was deployed with the 570-megapixel Dark Energy Camera on the NSF Víctor M. Blanco 4-meter Telescope at Cerro Tololo Inter-American Observatory in Chile.

The scheduler was created through the NSF-Simons Foundation AI Institute for the Sky and trained on 13 years of historical observations from the DOE-funded Dark Energy Survey. Instead of relying on hand-coded astronomy rules, the deep-learning model learned by predicting where the telescope should point next, comparing its choices with past human decisions and correcting errors over repeated training.

The team said the system has completed successful observing runs on the Blanco Telescope, with performance currently comparable to human schedulers. Researchers now aim to move beyond imitation by developing models that can test observing strategies humans may not consider, potentially helping future observatories and companion telescopes make better use of limited observing time.