TrajAir
CMU AirLab · 2021 · datasets.bot · datasets.bot page
One-liner. Real general-aviation aircraft trajectory dataset (111 days over 8 months) of ADS-B tracks plus METAR weather, recorded around the non-towered Pittsburgh-Butler airport (KBTP) for socially-aware trajectory prediction.
Setup
- Datasets / benchmarks: TrajAir is a general-aviation aircraft trajectory dataset collected by CMU's AirLab (Robotics Institute) around the non-towered Pittsburgh-Butler Regional Airport (ICAO: KBTP), a single-runway GA airport ~10 miles north of Pittsburgh, PA. Data was recorded with an on-site ADS-B receiver (1090 MHz / 978 MHz, plus TIS-B for non-equipped aircraft) over 8 months (18 Sept 2020 - 23 Apr 2021), yielding 111 valid days after excluding downtime, repairs, and no-traffic bad-weather days. Each scene contains aircraft trajectories transformed into an airport-centered inertial frame with x,y,z positions (km), aircraft IDs, and frame timestamps interpolated to 1 Hz via cubic interpolation. Weather context (wind direction/velocity, given as wx, wy components in m/s) is added post-hoc from METAR data produced by the KBTP AWOS system. The data is distributed as delimited text files organized into a full 111_days/ folder, four weekly folders (7days1-4), raw CSVs by collection date, and a processed 70/30 train/test split. It was released with the baseline socially-aware prediction method TrajAirNet (ICRA 2022). This is real physical manned-aircraft data, not simulated. License: CC-BY-4.0. Download: https://kilthub.cmu.edu/ndownloader/articles/14866251/versions/1.
- Hardware / simulator: Embodiment: drone. Environment: outdoor. Realness: physical.
Schema
Delimited text files of scenes; columns: Frame#, Aircraft ID, x, y, z (km, airport-centered inertial frame), wind components wx, wy (m/s); 1 Hz sampling. Organized as 111_days/, weekly folders 7days1-4, raw CSV by date, and processed 70/30 train/test split.
Links