sensor-fusion

What it does

Describes a specialized skill for multi-sensor fusion and state estimation using Kalman filtering and factor graph optimization. It enables implementing Extended Kalman Filter (EKF) for state estimation and configuring an Unscented Kalman Filter (UKF), along with robot_localization setup, IMU preintegration and bias estimation, GPS/RTK integration, wheel odometry fusion with slip compensation, visual odometry integration, outlier rejection, covariance tuning, and sensor delay compensation.

How it works

• Provides ROS2 robot_localization configuration for EKF/UKF, including two example YAML blocks: one for continuous odometry using IMU and wheel odometry, and another for global localization combining odometry and GPS. It specifies frames, frequencies, input topics, and which state elements to fuse, plus process and initial covariances.
• Describes a Two-EKF setup: one EKF for continuous odometry (odom frame) and another for global/localization (map/world frame) with specific fusion of IMU, odometry, and GPS data.
• Includes a NavSat transform node configuration to broadcast GPS in a usable frame, with parameters for delays, yaw, and transforms.
• Includes a Custom EKF implementation (Python) with a 12-state vector [x, y, z, roll, pitch, yaw, vx, vy, vz, wx, wy, wz], with predict step using a constant velocity model and Jacobian computation, update steps for IMU, odometry, and GPS measurements, and a protected GPS outlier rejection based on Mahalanobis distance.
• Provides an IMU Preintegration class for factor-graph optimization, handling delta_R, delta_v, delta_p, and covariance, with bias handling and Jacobian updates.
• Contains a Noise Covariance Tuning function template for autocalibration of Q and R through innovation analysis, including a skeleton for processing a data log and computing NEES over measurements.

When to use it

• Use when developing robust robot localization with multiple sensors (IMU, wheel odometry, GPS, visual data) in ROS2 environments.
• Use when you need EKF/UKF variants, IMU preintegration, and outlier rejection for GPS measurements.
• Use when configuring robot_localization with precise frame mappings and timing settings, and when you require tuning guidance for process and measurement noise covariances.

What it can touch

• robot_localization configuration through YAML parameters (ekf_filter_node, ekf_filter_node_odom, ekf_filter_node_map).
• navsat_transform_node parameters for GPS integration.
• Custom Python modules: RobotEKF (state estimation), IMUPreintegration (preintegration), and tune_noise_covariances function (noise calibration).
• Input topics such as /imu/data, /wheel_odom, /odometry/filtered, /gps/odom.

Caveats

• Requires ROS2 with robot_localization package and calibrated sensors (IMU, cameras, wheel encoders).
• Original text includes extensive YAML examples and Python code snippets; actual outcomes depend on real sensor data and proper tuning.
• GPS outlier rejection uses a fixed Mahalanobis distance threshold (5.0) as shown in the code.
• The IMU preintegration implementation provides a simplified covariance propagation and may need adaptation for production-grade factor graphs.