ROS 2 UKF sensor fusion SDK combining IMU, wheel encoders, and GPS with ECEF-native fusion, adaptive noise covariance, and zero manual tuning.
Key Takeaways
Fills a real ROS 2 gap: robot_localization has no ECEF GPS or IMU bias estimation; fuse has no RTK quality gating as of early 2026.
NCLT dataset benchmarks: FusionCore wins 5 of 6 sequences vs RL-EKF; RL-UKF diverges with NaN on all six sequences.
22D quaternion UKF state tracks gyro and accel biases; chi-squared outlier gating applied per-sensor with GPS fix-quality gating (GPS/DGPS/RTK_FLOAT/RTK_FIXED).
Supports dual antenna heading, lever arm correction with observability guard, ZUPT while stationary, and a ~/reset service that reinitializes without restarting the node.
Apache 2.0, native ROS 2 Jazzy (not a ROS 1 port), 39 tests passing, per-sensor SensorHealth diagnostics at 1 Hz.
Hacker News Comment Review
The one technical commenter pointed to WOLF as a competing pluggable sensor fusion framework worth evaluating before adopting FusionCore.
The README’s tone drew skepticism about AI authorship, which may affect trust in the benchmark methodology and claimed results.
The remaining discussion is thin; one commenter dismissed the project based on ROS aversion rather than evaluating the fusion logic.
Notable Comments
@jeremysalwen: flags WOLF framework as a promising pluggable alternative and calls the README “obviously AI generated.”
