China's AI-powered StarWhisper Telescope gains global recognition

A Chinese telescope system that uses artificial intelligence to plan and conduct observations has been listed in Stanford's AI Index Report 2026, highlighting the growing global recognition of China's efforts to integrate large models into front-line scientific discovery.

The StarWhisper Telescope, developed by the National Astronomical Observatories of the Chinese Academy of Sciences, is featured in the report's science chapter as a representative case of AI agents in physics, astronomy, chemistry, and materials sciences. The report was recently released by the university's Institute for Human-Centered Artificial Intelligence.

The telescope takes into account scientific objectives, equipment status and weather conditions, enabling AI agents to participate in task planning and observation execution. This marks a shift from using AI solely for data analysis to integrating it into the operation of complex instruments.

To support this transition, the National Astronomical Observatories has built a digital simulation environment for large-aperture research telescopes. The platform mimics optical tubes, mounts, cameras and sensors, allowing agents to test control strategies without occupying valuable observing time on real facilities.

The work received guidance from the academy's Panshi large-model team and technical support from Alibaba Cloud. Leveraging large-model-assisted development capabilities, the team completed the core simulation environment in a short period, with an estimated development cost of about 1,000 yuan ($149).

"The research paradigm is changing. AI's role in research is no longer limited to assisting with analysis; it is entering the research execution process itself," Li Yuyang, an expert with the National Astronomical Observatories' Artificial Intelligence Steering Committee, said.

"Researchers will focus more on formulating questions and setting objectives, while AI agents undertake observation execution, validation, and iteration," Li said.

The research team is also developing models to help agents decide both what to observe and how to observe it. A cross-survey aligned time-series model identifies very early supernova candidates from public datasets, while a short-term local weather model evaluates cloud cover, wind speed, and humidity to optimize observing plans.

These capabilities have been integrated into the Sitian Pathfinder — a Chinese sky-survey test bed — and its full prototype system. The agent framework has so far issued alerts for eight very early supernova candidates, with two of them triggering follow-up observations when weather conditions permitted.

This enabled researchers to submit observing requests in natural language. The agent then generates a plan, invokes control modules through standardized interfaces, and reports results. After each task, it adjusts future plans based on execution outcomes and expert feedback, distilling effective experience into reusable AI skills.

The simulation system can repeatedly execute observation tasks in a virtual environment while recording equipment status, environmental changes and decision-making processes. This creates high-quality training data for intelligent telescope control and lays a foundation for vision-language-action models tailored to astronomical observation.

The National Astronomical Observatories said that the broader goal is to move from AI-assisted research to AI-participatory research. In this emerging paradigm, scientists focus on asking questions and defining objectives, while AI agents handle observation execution, validation and iteration.

It also plans to release related models and tools on the academy's Panshi platform to encourage sharing across the global astronomical community. This effort reflects a wider shift in scientific research, where intelligent systems are becoming active participants in the discovery process.