navfusion

What it solves

navfusion targets one of the highest-cost autonomy failures: navigation stacks that degrade sharply when sensor timing or quality departs from nominal conditions. It provides a modular fusion backbone built for engineering teams that need predictable behavior, diagnosable failures, and reproducible validation outputs.

• Maintains stable state estimation during GNSS dropouts and delayed measurement bursts.
• Reduces integration friction for mixed-rate, asynchronous sensor pipelines.
• Supports recruiter-facing and safety review narratives with concrete diagnostics.

Engineering highlights

• Real-time-first event-driven fusion engine.
• Error-state EKF with quaternion attitude representation.
• IMU propagation plus GNSS position/velocity fusion.
• Asynchronous sensors with bounded out-of-order handling.
• Innovation gating and robust diagnostics for measurement integrity.
• Modular architecture for extension toward additional sensor modalities.

Reliability and validation

Reliability is treated as a first-class output, not a side effect. Validation pipelines are designed to expose estimator inconsistency early and to preserve reproducibility across benchmark runs.

• NIS/NEES consistency checks integrated in CI.
• Benchmark artifacts captured for post-run review and hiring-panel walkthroughs.
• Reproducible docs and demo outputs tied to versioned code paths.
• Diagnostics centered on innovation behavior and gate decisions.

Demo scenarios

1. GNSS dropout handling.
2. Out-of-order and delayed GNSS robustness.
3. GNSS outlier rejection with gating.

Links

Repo: github.com/leo-01000111/navfusion
Docs: www.leongorecki.eu/navfusion/

Next steps (planned)

Planned path includes an Unscented Kalman Filter (UKF) branch for stronger nonlinear handling, while preserving current benchmarking discipline and consistency diagnostics for fair A/B evaluation.