I built FusionCore after spending weeks fighting robot_localization on a mobile robot. The filter kept diverging, the docs said "tune your Q matrix" with zero guidance, and there was no way to know *why* it was failing. FusionCore is a 22-state UKF that fuses GPS, IMU, and wheel odometry for ROS 2. Three things I did differently: **Adaptive noise estimation**: the filter tunes its own measurement noise online from a sliding innovation window. You don't touch a Q matrix. **Mahalanobis outlier rejection**: bad GPS fixes and IMU spikes are detected and dropped before corrupting the state. On the NCLT outdoor driving dataset, this is the difference between 5.6 m ATE and total divergence. **ECEF-native GPS**: no UTM zone crossings, no ENU approximation errors at long range. Works anywhere on Earth. I benchmarked against robot_localization EKF on 6 sequences of the NCLT dataset (90 km of outdoor driving). FusionCore wins on 5 of 6, up to 12.9× better ATE. RL-UKF diverges to NaN on every sequence... known numerical instability. Apache 2.0, fully open source. If you're doing robot navigation, localization, or autonomous vehicles in ROS 2, I'd love your feedback.

FusionCore

ROS 2 UKF sensor fusion: GPS+IMU+wheels, zero manual tuning

ROS 2 UKF sensor fusion: GPS+IMU+wheels, zero manual tuning