Learning to Land Anywhere: Transferable Generative Models for Aircraft Trajectories

2025-11-06

AI Summary

Data scarcity at low-traffic airports hinders the application of generative models in air traffic management. Method: This paper proposes a transfer learning–based cross-airport aircraft landing trajectory generation framework, pioneering the integration of diffusion models and flow matching (including latent flow matching) architectures into aviation. It adopts a pretrain-fine-tune paradigm: a general-purpose trajectory generator is pretrained on data-rich airports (e.g., Dublin), then fine-tuned using only 5%–20% of local trajectory data from the target airport. Results: With just 20% of target-airport data, the transferred model matches the performance of a baseline trained on full data—substantially outperforming from-scratch training. Even with only 5% data, it maintains high-fidelity trajectory synthesis. The approach overcomes the key bottleneck in few-shot high-quality trajectory generation, enabling scalable and robust generative simulation and what-if analysis for data-constrained airports.

Technology Category

• Natural Language Processing: Generation
• Planning, Routing, and Scheduling: Planning with Language Models
• Computer Vision: Diffusion Models for Vision

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Web Mining and Content Analysis: Web data generation and simulation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Access to trajectory data is a key requirement for developing and validating Air Traffic Management (ATM) solutions, yet many secondary and regional airports face severe data scarcity. This limits the applicability of machine learning methods and the ability to perform large-scale simulations or"what-if"analyses. In this paper, we investigate whether generative models trained on data-rich airports can be efficiently adapted to data-scarce airports using transfer learning. We adapt state-of-the-art diffusion- and flow-matching-based architectures to the aviation domain and evaluate their transferability between Zurich (source) and Dublin (target) landing trajectory datasets. Models are pretrained on Zurich and fine-tuned on Dublin with varying amounts of local data, ranging from 0% to 100%. Results show that diffusion-based models achieve competitive performance with as little as 5% of the Dublin data and reach baseline-level performance around 20%, consistently outperforming models trained from scratch across metrics and visual inspections. Latent flow matching and latent diffusion models also benefit from pretraining, though with more variable gains, while flow matching models show weaker generalization. Despite challenges in capturing rare trajectory patterns, these findings demonstrate the potential of transfer learning to substantially reduce data requirements for trajectory generation in ATM, enabling realistic synthetic data generation even in environments with limited historical records.

Problem

• Overcoming data scarcity at regional airports for trajectory generation
• Adapting generative models from data-rich to data-scarce airports
• Reducing data requirements for air traffic management simulations

Innovation

• Transfer learning adapts generative models between airports
• Diffusion models achieve baseline performance with 20% data
• Generative models reduce data needs for trajectory generation