Sensor Drift

传感器漂移（零漂）Advanced

A sensor’s reading slowly shifting over time or temperature even when the true input hasn’t changed.

An ideal sensor should read zero with no input, but a real sensor always carries some offset, or bias, and that bias itself changes slowly with time, temperature, and each time the sensor is powered on — this is drift, also called zero drift or thermal drift. The most common case in robotics is IMU (inertial measurement unit) drift: orientation, velocity, and position are all obtained by integrating gyroscope and accelerometer readings, so a tiny bias keeps accumulating — a constant gyroscope bias makes velocity error grow with the square of time and position error grow with its cube, so relying on an IMU alone to estimate position quickly diverges. Six-axis force/torque sensors and joint torque sensors also drift, so they typically need to be re-zeroed before use. Countermeasures include calibration, temperature compensation, characterizing noise with Allan variance, and fusing with cameras, lidar, or GPS, with the bias itself estimated online as part of the state in a Kalman filter.

ExampleA humanoid robot standing perfectly still still sees its yaw angle, computed purely by integrating the IMU, slowly drift over time, which needs correcting by fusing in visual-inertial or legged odometry.

Also called: Zero Drift, Bias Drift, Thermal Drift, IMU Drift
Related: Inertial Measurement Unit · Kalman Filter · Allan Variance · Visual-Inertial Odometry · Six-Axis Force/Torque Sensor · State Estimation
Sources: Inertial measurement unit - Wikipedia