multi-sensor-fusion

Multi-sensor fusion combines wheel odometry, IMU, visual odometry, and lidar odometry into a unified state estimate using EKF or UKF algorithms. You configure each sensor source, tune per-source covariance, and set differential or absolute measurement modes to achieve accurate localization.

EKF and UKF are both supported fusion methods for state estimation. EKF uses linearization for sensor fusion, while UKF handles nonlinear motion profiles more accurately. Both allow per-sensor covariance tuning and outlier rejection to reduce drift and inconsistent pose updates.

ROS2 sensor fusion supports per-sensor configuration for absolute versus differential measurements. You define each sensor stream's mode, adjust covariance values, and apply a two-EKF drift-correction pattern to maintain robust pose accuracy across various indoor and outdoor motion profiles.

EKF fusion maintains localization accuracy during partial sensor dropout by integrating multiple sensor streams. Combining wheel odometry, IMU, and visual odometry allows the state estimator to compensate dynamically when one sensor source fails or provides inconsistent updates.

Localization drift occurs when sensor covariance is improperly tuned or measurement modes are mismatched. Applying a two-EKF drift-correction pattern with correct per-source covariance tuning for visual odometry and IMU data stabilizes the fused state estimate and reduces drift.

Multi-sensor fusion is applicable to both indoor mapping and outdoor robots. It supports high-rate localization for indoor environments and drift correction for outdoor robots by fusing lidar odometry, wheel odometry, and IMU data across various motion profiles.