Show HN: FusionCore: ROS 2 sensor fusion that outperforms robot_localization

· coding · Source ↗

TLDR

  • FusionCore is an Apache 2.0 ROS 2 SDK using a 22D quaternion UKF to fuse IMU, wheel odometry, and GPS, replacing robot_localization with ECEF-native GNSS, auto-tuning, and IMU bias estimation.

Key Takeaways

  • Beats robot_localization EKF on 5 of 6 NCLT dataset sequences; RL-UKF diverges with NaN on all six.
  • Core differentiators: ECEF-native single-node GPS fusion, RTK fix quality gating, dual antenna heading, lever arm correction, and ZUPT.
  • Self-tunes noise covariance via innovation sequence; no manual Q/R matrix work required for most deployments.
  • Per-sensor SensorHealth diagnostics, chi-squared outlier gating on all sensors, and a ~/reset service without node restart.
  • On one degraded-GPS sequence (2012-11-04), RL-EKF outperforms FusionCore because its lack of an outlier gate lets it self-correct immediately during prolonged GPS dropout.

Hacker News Comment Review

  • No substantive HN discussion yet.

Original | Discuss on HN