maneuver-detect

Open dataset, models, and benchmark for detecting orbital maneuvers from public TLE history.

maneuver-detect takes a satellite's public TLE history and returns a DataFrame of detected maneuvers — each with a detection epoch, a calibrated confidence, a maneuver type (in-track / cross-track / radial), and a Δv estimate. It ships a curated, reconstructable, labelled dataset; a classical reference detector, learned (BiLSTM and transformer) baselines, and a foundation-model (Chronos) baseline distributed through the Hugging Face Hub, all sharing a vis-viva / Gauss-variational Δv inversion; and a frozen, leak-free benchmark — with a public leaderboard — so a new detection method can be measured against prior work on the same splits.

Why

Detecting maneuvers from public TLEs is a long-running space-situational-awareness problem, but every paper rebuilds its own dataset, cleaning pipeline, detector, and evaluation — so results are not comparable and the data is rarely published. maneuver-detect provides the missing shared piece: an open, citable dataset and a reproducible benchmark, with a classical baseline every learned model must beat and a deterministic scorer that reproduces the published numbers.

What it is not

• Not a maneuver predictor — it detects maneuvers that have already happened, not future ones.
• Not a real-time or streaming pipeline — it is batch: a TLE history in, a maneuver DataFrame out.
• Not a new propagator or orbit-determination engine — it consumes SGP4 mean elements and the small inversions the Δv estimate needs.
• Not a general time-series-anomaly framework — the detectors are maneuver detectors on orbital element series.
• Not a cross-catalog correlation or object-association tool — it works one catalogued object's history at a time, not the multi-sensor tracking problem.
• Closed or commercial data is out of scope — only publicly redistributable data is used.

Explore

• Getting started — install, run the detector on a NORAD id, read the result.
• Dataset and label sources — what is in the dataset, how it is distributed, and each source's terms.
• Models and the Hub — the classical, learned, and foundation detectors, and how detect(model=…) loads a checkpoint from the Hugging Face Hub.
• Benchmark protocol — the splits, the matching rule, and the metric a method is scored on.
• Leaderboard — the public board on the frozen test split, and how to submit a method.
• Output schema and Δv inversion — the columns detect returns and the physics behind them.
• Design decisions — the frozen decision record (D1–D17).
• API reference — the full public surface.

See the changelog for released functionality.