Reproduce “Fast Surrogate Models for Adaptive Aircraft Trajectory Prediction in En route Airspace” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Nick Pepper, Marc Thomas, Zack Xuereb Conti
The LWPF system predicts aircraft climb and descent endpoints by adapting a bank of fitted LSSMs as six-second radar observations arrive, rather than repeatedly solving the detailed BADA performance model. On held-out UK radar data, it reduced time MAE versus KF-TP by 46.3% for climbs and 64.7% for descents, while adaptive BADA remained better than KF-TP for descent time.
Key insights
- Online adaptation is more valuable than simply reinitializing BADA: learning the aircraft’s evolving speed and performance profile substantially improves prediction of both time and distance.
- A lightweight linear surrogate can preserve BADA-like behavior while making sequential Monte Carlo practical at larger scale; it is approximately 5.26 times faster for climbs and 8.11 times faster for descents.
- Performance depends on the training-data coverage and aircraft dynamics: discontinuous BADA behavior, sparse aircraft types, and turboprop descent patterns expose cases where the particle filter is less reliable or where simpler baselines remain competitive.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.