Flow matching improves aircraft trajectory prediction accuracy around san francisco
FlowATC: Aircraft Trajectory Prediction via Flow Matching
Machine Learning
Summary
Predicting where planes will fly next is important for managing air traffic safely. The authors created a model that learns from past flight paths without needing extra labels, making it easier to train. Their model better matches real flight patterns near San Francisco Airport and can predict multiple possible future paths, helping estimate where planes might be in the airspace. This approach is more accurate than older models and works well even if data is less frequent.
What this means in practice
- •For air traffic controllers: Generate more precise forecasts of aircraft positions to assist in making safer traffic management decisions near busy airports.
- •For aviation software developers: Incorporate probabilistic aircraft position predictions into conflict-risk tools for advanced decision-support systems in airspace management.
Authors
Mathurin Petit, Emir Torun, Louis Brusset, Jordan Kam, Alexandre M. Bayen
Abstract
Building accurate decision-support tools for next-generation air traffic control requires robust trajectory prediction models. We present a flow-matching architecture trained exclusively on historical aircraft trajectories, with no route labels or chart supervision. Trained on 1.15 million Automatic Dependent Surveillance-Broadcast trajectory windows collected over the San Francisco Bay Area, the model generates aircraft trajectory distributions that closely match historical traffic, reproducing known airspace structure around San Francisco Airport such as the shape of SFO's published NIITE FOUR departure procedure. Our model is trained directly on the native, irregular ADS-B sampling interval. Trajectory prediction is cast as sequence inpainting using a block-causal Transformer that denoises future state tokens conditioned on the observed history using Conditional Flow Matching or Denoising Diffusion Probabilistic Models. We compare our architecture against constant-velocity, deterministic-Long Short Term Memory, and Conditional Variational Autoencoders baselines. At matched parameter count, CFM outperforms DDPM by 11-26% in minADE@20, and both generative objectives surpass the CVAE baseline by 31-41%. We further show that the error degrades gracefully with prediction horizon, and the architecture remains effective when retrained on temporally decimated feeds. Lastly, we sample $K$ independent completions, yielding spatial probabilistic occupancy estimates that can serve as input to downstream conflict-risk estimation.