IMU orientation drifts

1. Symptoms

• Yaw changes slowly with the robot stationary.
• Roll and pitch settle at a wrong value or slide.
• The drift rate changes as the IMU warms up.

2. Possible causes

• Gyro bias: a constant rate offset integrates into a linearly growing angle.
• Temperature: bias changes while warming up.
• No absolute reference for yaw: accelerometer fixes roll and pitch (gravity) but not yaw; a magnetometer, GNSS heading or vision is needed.
• Magnetometer disturbance from motor currents or steel (a corrupted yaw reference).
• Wrong units or axes into the filter (the Imu message uses m/s² and rad/s).
• Vibration biasing the accelerometer.

3. Measurements to collect

• Gyro output with the robot still for several minutes (mean = bias, standard deviation = noise).
• Drift rate in degrees per minute, before and after warm-up.
• Magnetometer field magnitude and direction with the motors off and on.
• The IMU covariance values and the filter’s inputs.

4. Diagnostic commands

Commands are not executed by the site’s automated tests (they need a ROS 2 Jazzy installation or the robot).

ros2 topic echo /imu/data --field angular_velocity --once
ros2 bag record /imu/data /imu/mag
# analyse offline: mean and std of angular_velocity while still

5. Step-by-step investigation

1. Record 5 minutes with the robot still and compute the mean of each gyro axis. A non-zero mean is the bias; its integral is the drift.
2. Repeat after warm-up to see temperature dependence.
3. Check units and axes: the message uses rad/s and m/s²; the frame follows REP 103 (x forward, y left, z up).
4. Check what stops yaw drift: is a magnetometer, GNSS heading or visual/LiDAR correction fused?
5. Check the magnetometer with motors on and off; a change means disturbance.
6. Check covariance: an all-zero covariance means unknown, and a first element of −1 means no estimate; the filter must not treat these as perfect.

6. Corrective actions

• Subtract the measured bias at start-up (stationary calibration) and let the filter estimate it online.
• Allow warm-up, or compensate temperature if the datasheet gives a coefficient.
• Add a yaw reference: magnetometer after calibration and away from currents, dual-antenna GNSS, or visual/LiDAR odometry.
• Mount the IMU rigidly and away from motors; use vibration isolation if needed.
• Correct units, frames and covariances.

7. Validation after the fix

• Stationary drift rate measured and below the requirement over the mission time.
• Yaw error after a loop or a long run is within the requirement with the reference sensors enabled.
• Results repeat after a cold start.

References

• sensor_msgs definitions (Jazzy): ros2/common_interfaces, read 2026-11-05. Imu: m/s² and rad/s; zero covariance means unknown; first element −1 means no estimate.
• REP 103: rep-0103.rst.
• Related guide: inertial and positioning sensors guide.