Rubin and Roman sky surveys
Rubin's 3.2-gigapixel camera began a 10-year repeated survey of the southern sky in June 2026, and NASA's Roman telescope launched on 30 August 2026.
Open in the interactive tree →Rubin Observatory in Chile (8.4 m mirror) images the whole visible sky every few nights to find moving and changing objects such as asteroids, supernovae and variable stars. Roman, with a field of view about 100 times Hubble's, will map dark energy and find planets by microlensing.
As of October 2026
Rubin released its first-light images on 23 June 2025, and the 10-year Legacy Survey of Space and Time began on the night of 29 June 2026, with up to 10 million alerts per night and about 20 billion galaxies and 17 billion stars to be catalogued. Roman lifted off on a Falcon Heavy on 30 August 2026 at 7:26 am EDT, is on a three-month journey to the Sun-Earth L2 point, and NASA expects to release its first images by early 2027.
Open steps
- Photo-z for dark-energy High AI leverageTrain photo-z models on multi-band imaging to achieve 1% photo-z accuracy for billions of galaxies.
- Supernova classification High AI leverageBuild real-time classifiers for supernova types from LSST light curves to prioritise spectroscopic follow-up.
- Habitable-zone direct imaging Low AI leverageUse Roman wide-field survey to directly image and characterise exoplanets in nearby habitable zones.
Where AI could help
High AI leverage. Millions of alerts a night cannot be handled by people; machine-learning brokers and survey analysis are required parts of the pipeline.
- Classify transients in seconds so follow-up telescopes target the right objects
- Detect anomalies and unknown object types across billions of catalogue entries
- Estimate photometric redshifts and shapes for about 20 billion galaxies
- Link faint moving detections to find new asteroids and comets
Shown so far
- In February 2026, Rubin sent its first alerts, 800,000 in one night; seven brokers use machine learning to filter, sort and classify them, and up to about 7 million a night are expected. source
- In April 2024 (revised November 2024), a Fink-team paper benchmarked classifiers on simulated LSST alerts, with a deep-learning model reaching at least 93% precision for all classes but one (about 83%). source
Prerequisites
- The telescope1608
- Integrated Circuit1958-1959
- CCD image sensor1969Rubin's 3.2-gigapixel camera is a mosaic of CCDs
- Deep Learning2012
- Gaia: 3D Map of the Milky Way2018–2022Rubin's sky positions are calibrated against Gaia's star catalogue