ASTRONOMERS at the University of Warwick have uncovered more than a hundred previously hidden worlds by applying a new artificial intelligence tool to data from Transiting Exoplanet Survey Satellite (TESS), marking a significant advance in the search for planets beyond the solar system.
Using a system called RAVEN, or Rapid Analysis and Verification of Exoplanets, the research team validated 118 exoplanets, including 31 that had not been identified before. The findings were published in Monthly Notices of the Royal Astronomical Society and demonstrate how machine learning is accelerating the ability of scientists to extract meaningful discoveries from vast astronomical datasets.
The TESS mission surveys the sky by monitoring the brightness of stars and detecting small dips caused when a planet crosses in front of its host star, a method known as transit photometry. However, these signals are frequently contaminated by false positives, such as binary star systems or stellar variability, which can mimic the presence of a planet. To address this, the Warwick team trained RAVEN using hundreds of thousands of simulated scenarios, allowing the system to distinguish genuine planetary signals from noise with a high degree of accuracy.
The AI pipeline was then applied to observations of more than 2.2 million stars collected during the first four years of TESS operations. In addition to confirming 118 planets, the system flagged more than 2,000 high-quality candidates and identified nearly 1,000 new potential planets that had not been previously catalogued. These results underscore the scale of undiscovered worlds still embedded in existing datasets and the growing role of AI in uncovering them.
The study focused on short-period planets, or those that orbit their stars in less than 16 days. Among the discoveries were ultra-short-period planets that complete an orbit in under 24 hours, exposing them to extreme temperatures and radiation. The analysis also provided one of the most precise measurements to date of the so-called Neptunian Desert, a region where Neptune-sized planets are rarely found in close orbits around Sun-like stars. The team determined that such planets occur around only about 0.08 percent of these stars, reinforcing the idea that certain planetary configurations are inherently uncommon.
Several multi-planet systems were also identified, offering further opportunities to study how planetary systems form and evolve. According to Dr. David Armstrong, the use of AI enables researchers to process enormous datasets in a consistent and objective manner, producing not just a list of candidates but a reliable statistical picture of planetary populations.
In a related analysis, the researchers used the RAVEN dataset to estimate that roughly 9 to 10 percent of Sun-like stars host at least one close-in planet. This figure is broadly consistent with earlier findings from the Kepler Space Telescope, but the new approach significantly reduces uncertainty, improving precision by an order of magnitude.
The implications extend beyond cataloguing new worlds. By producing a rigorously validated list of exoplanets, the study provides a set of high-confidence targets for follow-up observations. Instruments such as the James Webb Space Telescope and the planned PLATO mission are expected to use these targets to probe planetary atmospheres, compositions and potential habitability.
The findings highlight a shift in modern astronomy, where the challenge is no longer the lack of data but the ability to interpret it. With missions like TESS generating enormous volumes of observations, tools like RAVEN are becoming essential in transforming raw data into scientific discovery, offering a clearer view of how common planetary systems are across the galaxy.
