Research · 2026
flightMight: In-Firmware GPS Spoofing Detection on a Custom Flight Controller
Manuscript in preparation
Abstract
GPS spoofing detectors for small drones usually run on a companion computer that reads telemetry, so they depend on a link that can be lost, delayed or attacked. This work moves a five-feature, persistence-gated spoofing detector into the flight firmware itself.
Running as an ArduPilot Lua script on AZKAREL, a custom STM32H743 flight controller, the detector reads the EKF’s normalised innovations directly and combines them with position, velocity, acceleration and GNSS-quality residuals into an anomaly score that must persist before an alarm is raised. The work also covers the board bring-up needed to host it: an ArduPilot hardware definition, sensor calibration tooling and a MuJoCo physics bridge for software-in-the-loop testing.
Contributions
- A firmware-resident detector that reads EKF innovations through the flight controller’s own estimator instead of MAVLink telemetry.
- ArduPilot board support for AZKAREL: a hardware definition, bench calibration tools and a MuJoCo SITL bridge.
- A frame-of-reference correction that compares GPS kinematic acceleration with gravity-compensated IMU acceleration, removing false alarms that had triggered on every clean flight.
Results in simulation
| Clean flight, 252 s | No false positives |
|---|---|
| Gradual 0.08 m/s spoofing ramp | Suspicious at 1.70 s, confirmed at 5.00 s |
| Raw residual at confirmation | 0.10 m, against 10.3 m of actual displacement |
Validated in software-in-the-loop simulation. Hardware validation is pending, and detector thresholds will be recalibrated against real sensor noise.