robotics//sensor//IMU

An IMU (inertial measurement unit) is a sensor package that measures its own linear acceleration and rotation rate along three axes, with an accelerometer and a gyroscope per axis, and it is the fast heart of every estimator that tracks a moving body: the attitude of a drone, the motion of a phone, the inertial navigation of an aircraft. In drones and phones it is a MEMS chip costing a few euros that delivers data at 1 to 8 kHz over SPI; inside, it samples at several kilohertz and filters before handing over each value, so that motor vibration does not fold into the band of the attitude loop with little delay added. Packages that add a magnetometer are often called nine-axis, and with an attitude estimator on board the package is sold as an AHRS.

An IMU (inertial measurement unit) is a sensor package that measures its own linear acceleration and rotation rate along three axes, with an accelerometer and a gyroscope per axis, and it is the fast heart of every estimator that tracks a moving body: the attitude of a drone, the motion of a phone, the inertial navigation of an aircraft. In drones and phones it is a MEMS chip costing a few euros that delivers data at 1 to 8 kHz over SPI; inside, it samples at several kilohertz and filters before handing over each value, so that motor vibration does not fold into the band of the attitude loop with little delay added. Packages that add a magnetometer are often called nine-axis, and with an attitude estimator on board the package is sold as an AHRS.

The two members share what makes the package useful and what makes it hard. Both measure in the body's own frame, so their axes must be calibrated for scale, bias and misalignment as a 3×3 matrix plus an offset (sensor calibration); both carry a bias that moves with temperature and time (sensor bias); both are integrated to give attitude, velocity and position, so every one of their errors accumulates (dead reckoning). The estimator uses them as its prediction step at hundreds of hertz and lets the slower sensors correct it.

The grade decides how long the integration can be trusted. IMU grades are separated by bias stability, which spans four orders of magnitude: a few to tens of degrees per hour for consumer MEMS, around 1 °/h for tactical grade and 0.01 °/h for navigation grade (fibre-optic and ring-laser gyroscopes), with prices from euros to tens of thousands (Allan variance). A tactical IMU bridges minutes without GPS where a phone's bridges seconds.

Some quantities it can never see. Gravity does not change when the body turns about the vertical, so an IMU alone cannot observe its heading, nor its absolute position: those need another sensor, never a better filter (observability). In a hover even an accelerometer bias and a small tilt error give the same reading, and only manoeuvring separates them.

Its mounting is part of its specification. On a flight controller it sits on rubber dampers, a mechanical low-pass filter against propeller vibration, and its data travel with a timestamp from the chip or the interrupt that read it (time synchronization).