Qwen Office Helps NAOC Team Build Telescope Simulation in Three Days

Transient events such as supernova explosions and gamma-ray bursts are important for studying cosmic evolution and extreme physical processes, but their timing is hard to predict. Research-grade telescope resources are limited, and real equipment cannot support extensive trial and error. Before formal observations, teams usually need to build a telescope simulation system to model the full observation process and verify a plan before using real instruments. Such systems previously depended on software vendors, took about three months to develop and cost tens of thousands of yuan, according to Lei Feng News.

Qwen Office helped the NAOC team integrate telescope components, sensor status and observation environment information into standard MCP interfaces for the agent to monitor and call. In the simulation, a cross-survey alignment timing model can identify very early supernova candidates from public data from multiple surveys, telling the agent what to look at. A short-term local weather prediction model evaluates observing conditions by combining real-time parameters such as cloud cover, wind speed and humidity, helping the agent understand how to look. The agent can then combine scientific target priority, real-time telescope status and target observability windows to autonomously complete a closed loop of state perception, task planning, plan generation and process verification.

The agent framework has been connected to the Sitian Pathfinder and the Sitian prototype, according to the report. After researchers submit scientific observation requirements, the agent generates an observation plan based on target priority, weather and equipment status, calls the relevant control modules through different MCP interfaces, and feeds observation results back to researchers. After a task ends, it can replan subsequent tasks based on execution results and expert feedback, and turn effective experience into reusable Skills. So far, the agent has warned of eight very early supernova candidates, and two triggered follow-up observations when weather conditions allowed.

The effort is also an exploration of a new way to organize and execute astronomical research, beyond providing a simulation system for a single telescope. In the past, telescope automation systems mainly ran according to preset processes; when weather, equipment status or scientific goals changed, researchers still had to make judgments and adjustments manually. As large models, agents, scientific models and telescope control systems become connected, AI can dynamically plan, call tools, verify plans and adjust observation strategies based on real-time environments and scientific goals. The simulation system can also repeatedly execute observation tasks in a virtual environment, continuously recording equipment status, environmental changes, decision processes and execution results. This creates high-quality training data for intelligent telescope control and supports further training of a telescope control model, or VLA, and higher-level intelligent observation on real scientific facilities in the future.

Li Yuyang, a doctor and expert on the National Astronomical Observatories' AI Promotion Committee, said the research paradigm is changing. "AI's role in research is no longer only auxiliary analysis, but further entering the research execution process itself. Researchers are more responsible for proposing questions and setting goals, while agents undertake observation execution, verification and iteration," Li said.

Editor's Summary

Qwen Office helped a National Astronomical Observatories team build a research-grade telescope simulation system in three days for under 1,000 yuan, replacing a process that previously took about three months and cost tens of thousands of yuan. An AI agent connected to the Sitian telescope framework has flagged eight very early supernova candidates, two of which triggered follow-up observations when weather allowed. The project also tests a model in which AI agents handle observation planning and execution while researchers set scientific goals.