Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

AirTrafficGen: Configurable Air Traffic Scenario Generation with Large Language Models

Created by
  • Haebom

Author

Dewi Sid William Gould, George De Ath, Ben Carvell, Nick Pepper

Outline

To address the challenges and time-consuming manual design of Air Traffic Control (ATC) training scenarios, this paper proposes $\texttt{AirTrafficGen}$, a novel end-to-end scenario generation system leveraging Large-Scale Language Models (LLMs). $\texttt{AirTrafficGen}$ uses graph-based representations to encode sector topology, including airspace geometry, airways, and fixed points, into a format that LLMs can process. Using state-of-the-art models such as Gemini 2.5 Pro, OpenAI o3, GPT-oss-120b, and GPT-5, it generates high-traffic scenarios resembling real-world operations, with engineered prompts providing fine-grained control over the presence, type, and location of interactions. Furthermore, it demonstrates the feasibility of iterative improvement, correcting erroneous scenarios based on simple textual feedback. This system offers a scalable alternative to manual scenario design and contributes to the expansion of the quantity and diversity of ATC training and validation simulations. More broadly, it demonstrates the potential of LLMs for complex planning in safety-critical areas.

Takeaways, Limitations

•
Takeaways:
◦
Leveraging LLM to automate and streamline ATC training scenario creation
◦
Ability to create diverse and realistic high-traffic scenarios
◦
Fine-grained scenario control and iterative improvement possible
◦
Presenting the potential of LLM in safety-critical areas
◦
Save time and resources over manual scenario design
•
Limitations:
◦
The performance of the presented model may depend on the specific LLM. Performance verification on other LLMs is required.
◦
Additional verification is needed to ensure perfect alignment with the actual operating environment.
◦
A method to ensure reliability and safety of LLM output is needed.
◦
Consideration must be given to system stability and maintenance during long-term operation.
◦
As these are preliminary research results, more extensive experiments and verification are needed.
👍
Made with Slashpage