Sensor Measurement Models in Project Aria Devices

This page provides an overview of how Project Aria device sensor measurements are modeled for IMU, magnetometer, barometer and audio.

IMUs

For IMUs, we employ an affine model where the value from the readout of accelerometer $s_a$ or gyroscope $s_g$, is compensated to obtain a "real" acceleration $a$ and angular velocity $\omega$ by

$$a = M_a^{-1}(s_a - b_a) \qquad \omega = M_g^{-1}(s_g - b_g)$$

$M_a$ and $M_g$ are assumed to be upper triangular so that there is no global rotation from the imu body frame to the accelerometer frame.

Inversely, we can simulate the sensor read-out from acceleration or angular velocity by

$$s_a = M_a a + b_a \qquad s_g = M_g \omega + b_g$$

When the read-out signal exceeds a threshold, the signal saturates. Saturation limits are sensor dependent and referenced in the following table for accelerometer and gyrometers.

	accel-left	accel-right	gyro-left	gyro-right
saturation	8g	16g	2000	2000

Magnetometer, barometer and audio

Similar to the IMU rectification model, the sensor readouts for magnetometer, barometer, and audio data are modeled as linear to the real $r$ (magnetic field, air pressure and sound intensity).

Audio specifically is bias only.