IMU vs. AHRS: what is the difference and when do you need which?

01

What does an IMU actually output?

An inertial measurement unit contains three orthogonal rate gyroscopes and three orthogonal accelerometers. It measures angular velocity and linear acceleration, nothing more (Woodman, Cambridge UCAM-CL-TR-696). Many industrial IMUs output these values already integrated over one sample as delta angle and delta velocity, which is what a strapdown navigation algorithm needs. The Epson M-G370PDG0, for example, offers delta angle and delta velocity output but no attitude output (datasheet Rev 1.2).

Raw IMU data is the right choice when your system already runs its own estimator: a GNSS/INS filter, a SLAM stack, or a robot localization package that fuses wheel odometry, lidar and inertial data. Then the IMU's job is to deliver clean, well-specified rates and accelerations, and the datasheet values that matter are gyro bias instability, angle random walk and velocity random walk.

02

What does an AHRS add?

An attitude and heading reference system uses gyroscopes and accelerometers, usually together with magnetometers, and fuses them with a Kalman or complementary filter into an orientation estimate (Justa et al., Sensors 2020). Integrating the gyro gives a fast, smooth orientation that drifts. The accelerometer measures gravity, which gives an absolute but noisy reference for roll and pitch. The filter uses the second to pull the first back.

Two physical limits follow from that. First, gravity contains no information about rotation around the vertical axis, so the accelerometer cannot correct yaw (Wondosen et al., Sensors 2021). Heading needs a magnetometer, a dual-antenna GNSS or another external reference. Second, the accelerometer cannot separate gravity from the vehicle's own acceleration. During braking, cornering or vibration the tilt reference is contaminated (Justa et al.), and magnetometers are disturbed by local magnetic anomalies in their surroundings (Wondosen et al.). How well an AHRS handles both is a property of its filter, not of the sensor alone.

03

How does this compare with an INS?

An inertial navigation system goes one step further. It subtracts gravity, integrates acceleration twice and tracks position and velocity from a known start (Woodman). That makes attitude errors expensive: a residual tilt leaks gravity into the horizontal axes. Woodman's example shows a tilt error of 0.05° producing a horizontal position error of 7.7 m after 30 s.

04

Which Epson IMUs provide attitude output on board?

Two models in the Epson PDG series compute attitude internally: the M-G330PDG0 and the M-G366PDG0. Both output inclination, Euler angles or quaternions from their gyros and accelerometers. Neither has a magnetometer, and Epson states that yaw drift is not compensated. In practice that gives gravity-referenced roll and pitch plus a free-running yaw, which is what many levelling and pointing tasks need.

All values typical, from the current Epson datasheets (Rev 1.2) and product pages:

Dynamic accuracy is specified up to 100 °/s in Epson's motion profile "modeA", measured from a stationary start. All three share the 24 × 24 × 10 mm housing, about 10 g, and SPI or UART interfaces. On both the M-G330PDG0 and the M-G366PDG0, attitude output cannot be combined with the external trigger input, which matters if you synchronise the IMU to a camera or lidar.

05

How fast does yaw drift without a heading reference?

A simple estimate helps to judge whether free-running yaw is good enough. If the residual gyro bias stays at the bias-instability level and angle random walk adds on top, the yaw error after time t is roughly bias × t plus ARW × √t. This is our own simplified calculation with typical values at constant temperature; it ignores turn-on bias, temperature drift and scale factor errors.

For a gimbal that is re-referenced every few minutes, a fraction of a degree is often acceptable. For a machine that must hold heading for an hour without GNSS, the yaw budget decides the sensor, and heading aiding becomes part of the design. The IMU Selection Guide contains a drift calculator for your own durations.

06

When do you need which?

1. 01Your software already fuses several sensors (GNSS/INS, SLAM, ROS): use raw IMU data and pick by bias instability and random walk. Of the three models above, the M-G370PDG0 has the lowest bias instability.
2. 02You need roll and pitch for levelling, tilt compensation or stabilisation, and want to avoid writing a filter: use on-board attitude output (M-G330PDG0 or M-G366PDG0) and check the dynamic accuracy against your motion profile.
3. 03You need absolute heading: neither an IMU nor a magnetometer-free attitude output solves this alone. Plan a heading reference (dual-antenna GNSS, magnetometer in a clean magnetic environment, or map-based correction).
4. 04You need position through GNSS outages: you need an INS, so attitude errors and their effect on position belong in the error budget.