px4-conservation-poc

PX4 flight anomaly detection via conservation spectral analysis — detect attitude failures, GPS glitches, and motor failures from EKF2 state sequences.

Proof of concept applying the conservation spectral framework to PX4 autopilot data. Simulates 16-dimensional flight states (quaternion + velocity + position + gyro + accel), injects three types of anomalies, and detects them through conservation ratio drops in sliding-window spectral analysis.

What This Gives You

• Flight state simulator — synthetic PX4 EKF2 16D state sequences with smooth SLERP interpolation
• Three anomaly types — sudden attitude change, GPS glitch, motor failure
• Conservation-based detection — sliding window Laplacian, track CR drops
• Spectral fingerprinting — compare healthy vs unhealthy flight signatures
• Threshold comparison — conservation detection vs simple statistical thresholding
• Publication-quality visualization — 3D trajectories, CR timeseries, spectral fingerprints

Quick Start

pip install numpy scipy matplotlib scikit-learn
python demo.py

Outputs go to output/ with PNG plots.

Architecture

simulator.py   # FlightState + synthetic flight generation + anomaly injection
detector.py    # Conservation-based anomaly detection pipeline
visualize.py   # Professional dark-themed plots
demo.py        # End-to-end demo pipeline

How It Fits

Part of the SuperInstance ecosystem:

• conservation-spectral-python — Core SDK
• anomaly-atlas — Unified anomaly detection across domains
• px4-conservation-poc — PX4-specific proof of concept (this repo)

License

MIT