The telescope developed by the National Astronomical Observatories of China (NAOC) under the Chinese Academy of Sciences has been recognized in Stanford University's AI Index for 2026 report as a prime example of using artificial intelligence in fields such as physics, astronomy, chemistry, and materials science.
The Xingyu Telescope was created jointly by the NAOC's 'Milky Way Three-Dimensional Imaging Structure' group and the Xinlun Observational Base. It represents a new observation system with embodied intelligence in astronomy.
Unlike standard AI applications, which are typically used to analyze data after observations are complete, the Xingyu Telescope deeply integrates large models and AI agents directly into the telescope's control system. This allows AI to participate throughout the entire astronomical observation process, fundamentally changing the traditional approach to scientific research in this area.
Research telescopes have complex structures, and the quality of observations can depend on numerous factors, including equipment condition, weather conditions, and sensor parameters. Therefore, traditional methods rely heavily on manual control.
To solve this problem, the research group developed an inexpensive digital simulation system for telescopes. This system allows for the modeling of various observation scenarios, testing control strategies, and training AI agents without the need to spend time on real observations. The total development cost can be in the thousands of yuan (approximately 149 US dollars), significantly lowering the entry barrier for creating intelligent astronomical equipment and offering a new method for developing simulation systems for scientific objects.
The intelligent system has already successfully detected early warnings for eight supernova candidates at their earliest stages, two of which underwent subsequent observations. The AI agent is also capable of automatically saving acquired experience and iteratively improving observation strategies, thereby constantly increasing measurement accuracy.
