Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Amelia: A Large Dataset and Benchmark for Airport Surface Movement Forecasting

Ingrid Navarro, Pablo Ortega-Kral, Jay Patrikar, Haichuan Wang, Alonso Cano, Zelin Ye, Jong Hoon Park, Sebastian Scherer, Jean Oh

TL;DR

Amelia tackles rising demand and safety concerns in the U.S. National Airspace System by enabling large-scale data-driven forecasting of airport surface movements. The authors introduce Amelia-42, a SWIM-derived trajectory dataset across 42 airports, plus Amelia42-Mini, Amelia10-Bench, and Amelia-TF, a transformer-based baseline, all available as open-source. They define ego-centric long-horizon scenes (T=60s with history H=10s and forecast F=50s) and provide a multi-airport benchmark that supports in-domain and cross-domain evaluation. The results demonstrate improved cross-airport generalization with richer seen-environment diversity and illustrate the feasibility of long-horizon surface trajectory forecasting for safety-critical aviation applications, underscoring the practical impact of open data and scalable models.

Abstract

Demand for air travel is rising, straining existing aviation infrastructure. In the US, more than 90% of airport control towers are understaffed, falling short of FAA and union standards. This, in part, has contributed to an uptick in near-misses and safety-critical events, highlighting the need for advancements in air traffic management technologies to ensure safe and efficient operations. Data-driven predictive models for terminal airspace show potential to address these challenges; however, the lack of large-scale surface movement datasets in the public domain has hindered the development of scalable and generalizable approaches. To address this, we introduce Amelia-42, a first-of-its-kind large collection of raw airport surface movement reports streamed through the FAA's System Wide Information Management (SWIM) Program, comprising over two years of trajectory data (~9.19 TB) across 42 US airports. We open-source tools to process this data into clean tabular position reports. We release Amelia42-Mini, a 15-day sample per airport, fully processed data on HuggingFace for ease of use. We also present a trajectory forecasting benchmark consisting of Amelia10-Bench, an accessible experiment family using 292 days from 10 airports, as well as Amelia-TF, a transformer-based baseline for multi-agent trajectory forecasting. All resources are available at our website: https://ameliacmu.github.io and https://huggingface.co/AmeliaCMU.

Amelia: A Large Dataset and Benchmark for Airport Surface Movement Forecasting

TL;DR

Amelia tackles rising demand and safety concerns in the U.S. National Airspace System by enabling large-scale data-driven forecasting of airport surface movements. The authors introduce Amelia-42, a SWIM-derived trajectory dataset across 42 airports, plus Amelia42-Mini, Amelia10-Bench, and Amelia-TF, a transformer-based baseline, all available as open-source. They define ego-centric long-horizon scenes (T=60s with history H=10s and forecast F=50s) and provide a multi-airport benchmark that supports in-domain and cross-domain evaluation. The results demonstrate improved cross-airport generalization with richer seen-environment diversity and illustrate the feasibility of long-horizon surface trajectory forecasting for safety-critical aviation applications, underscoring the practical impact of open data and scalable models.

Abstract

Demand for air travel is rising, straining existing aviation infrastructure. In the US, more than 90% of airport control towers are understaffed, falling short of FAA and union standards. This, in part, has contributed to an uptick in near-misses and safety-critical events, highlighting the need for advancements in air traffic management technologies to ensure safe and efficient operations. Data-driven predictive models for terminal airspace show potential to address these challenges; however, the lack of large-scale surface movement datasets in the public domain has hindered the development of scalable and generalizable approaches. To address this, we introduce Amelia-42, a first-of-its-kind large collection of raw airport surface movement reports streamed through the FAA's System Wide Information Management (SWIM) Program, comprising over two years of trajectory data (~9.19 TB) across 42 US airports. We open-source tools to process this data into clean tabular position reports. We release Amelia42-Mini, a 15-day sample per airport, fully processed data on HuggingFace for ease of use. We also present a trajectory forecasting benchmark consisting of Amelia10-Bench, an accessible experiment family using 292 days from 10 airports, as well as Amelia-TF, a transformer-based baseline for multi-agent trajectory forecasting. All resources are available at our website: https://ameliacmu.github.io and https://huggingface.co/AmeliaCMU.
Paper Structure (60 sections, 19 figures, 9 tables)

This paper contains 60 sections, 19 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Trajectory Forecasting Datasets in Aviation.
  4. Trajectory Forecasting Models in Aviation.
  5. Dataset Collection and Processing
  6. Trajectory Data
  7. Raw Data.
  8. Processed Data.
  9. Map Data
  10. Dataset Analysis
  11. Trajectory Data Analysis.
  12. Map Data Analysis.
  13. A Trajectory Forecasting Benchmark
  14. Task Setup
  15. Scenario Representation.
  16. ...and 45 more sections

Figures (19)

  • Figure 1: The Amelia-42 Dataset. Total aircraft count per airport over a span of 15 days. We also include examples of airport scenes across multiple airports, showing diverse behaviors and interactions.
  • Figure 2: The Amelia data pipeline. A) Raw position reports from the FAA's SWIM Terminal Data Distribution System are continuously logged and released as Amelia-42, from December 2nd, 2022, to the present. B) Airport-specific geofences are defined to delimit movement areas as well as take-off and landing extensions to runways. C) Data within the geo-fence is processed into clean tabular 1-Hz position reports. D) As additional context, semantic routing graphs are created for each airport.
  • Figure 3: Analysis of the Amelia42-Mini subset. A) Shows the total number of agents, per type (Aircraft, Unknown, Vehicle). B) Shows an airport's heatmap, representing the activity frequency per region in the airport (darker, low frequency; brighter, high frequency). C) Shows the distribution of zone types (Taxiway, Runway, Hold-short Line) per airport's map.
  • Figure 4: Analysis of the Amelia10-Bench data. For each airport, we overlay a month's worth of processed trajectories spanning beyond runway limits to capture landing and take-off rolls. We also show each airport's crowdedness per hour of day.
  • Figure 5: An Overview of Amelia-TF.
  • ...and 14 more figures