IMU

This page covers the hardware side of the IMU stack:

sensors used across platforms
supported output families
IMU-to-camera alignment
factory calibration
noise characterization

For stream access and API usage, see the IMU node reference. For runnable pipelines, use the IMU examples in DepthAI.

All Kickstarter-backed OAK-D-Lite cameras do not have an onboard IMU.

IMU models

BNO08X family

9-axis sensor hub used on RVC2 devices, including raw and fused orientation outputs.

BMI270

6-axis accelerometer + gyroscope IMU used on RVC2 platforms as a raw inertial sensor path.

LSM6DSV

6-axis RVC4 IMU with accelerometer, gyroscope, and internal derived outputs.

AK09919

3-axis RVC4 companion magnetometer used when magnetic reference is needed.

Capability overview

Sensor	Axes	Physical sensors	Magnetometer	Sensor-side derived outputs	Absolute heading on-sensor
BNO08X	9	Accelerometer, gyroscope, magnetometer	Yes	Rotation vector, game rotation vector, geomagnetic rotation vector, gravity, linear acceleration	Yes
BMI270	6	Accelerometer, gyroscope	No	None exposed as sensor-side fused orientation outputs	No
LSM6DSV	6	Accelerometer, gyroscope	No	Game rotation vector, gravity vector, gyroscope bias	No
AK09919	3	Magnetometer	Yes	None	No

Output types

A common source of confusion is treating all IMU outputs as if they were generated in the same way. In practice, Luxonis devices expose multiple classes of IMU data, and not all of them are produced by the same processing path.

Raw

These are the direct sensor outputs as reported by the IMU itself, such as accelerometer, gyroscope, or magnetometer samples in the sensor's native frame.Raw outputs are useful when users want full control over downstream processing, for example in custom filtering, SLAM, VIO, or host-side sensor fusion pipelines.

Uncalibrated

Uncalibrated outputs are constructed from the raw sensor streams after applying the frame rotation needed to align them to the Luxonis camera coordinate convention:

+X right
+Y down
+Z forward

These streams are therefore still based on raw IMU measurements, but expressed in the expected device-aligned coordinate system.No factory intrinsic correction is applied at this stage.

Calibrated

Calibrated outputs are constructed from the raw sensor streams after applying:

frame rotation into the Luxonis RDF coordinate system
factory intrinsic calibration terms
compensation for deterministic sensor errors such as bias, scale error, shear / cross-axis terms, and assembly-related imperfections including small mounting or soldering deviations between the IMU and PCB

These are the recommended outputs when users want motion data that has already been corrected for known hardware imperfections.

Sensor-provided derived outputs

Some IMUs provide higher-level outputs directly from the sensor's internal firmware or fusion engine. Luxonis forwards these outputs as provided by the sensor.These include:

gravity vector
linear acceleration / gravity-compensated acceleration
rotation vector / quaternion outputs

For these outputs, Luxonis does not manually reconstruct or re-calibrate them from raw data. They are taken directly from the IMU's internal processing.

IMU reference frames

Each IMU package has its own native sensor axes defined by the vendor datasheet. These axes do not necessarily match the camera frame used by Luxonis image pipelines.Luxonis aligns IMU data to the camera frame using IMU extrinsics stored in calibration.

imuExtrinsics

imuExtrinsics is the rigid transform between the IMU frame and a camera frame.In practice:

for vectors such as acceleration and angular velocity, the rotation part of the transform is used
for full multi-sensor geometry, the full 4x4 transform is used

In practice, a vector measured in the IMU frame is rotated into the camera frame using the rotation component of imuExtrinsics.

Factory IMU calibration

Luxonis performs factory calibration of the IMU stack to compensate deterministic sensor errors, including:

bias
scale factor error
cross-axis / shear terms
assembly-related alignment imperfections

This is different from imuExtrinsics: intrinsic IMU calibration corrects the sensor itself, while extrinsic calibration relates the IMU frame to the camera frame.

Noise characterization

In addition to deterministic calibration, Luxonis characterizes the stochastic behavior of supported IMU models using noise parameters. These are useful for:

VIO
EKF / UKF pipelines
SLAM
simulation
covariance tuning
tools such as Kalibr

These parameters do not correct a fixed hardware error. Instead, they describe measurement noise and bias drift over time.Different tools use different names for similar IMU noise terms. The table below maps common IMU terminology to Kalibr naming.

Concept	Common IMU / Allan term	Kalibr name	Luxonis documentation name
Accelerometer white noise	VRW	`accelerometer_noise_density`	`accel_noise_density`
Accelerometer bias drift	Bias random walk	`accelerometer_random_walk`	`accel_bias_random_walk`
Accelerometer bias stability	Bias instability	-	`accel_bias_instability`
Gyroscope white noise	ARW	`gyroscope_noise_density`	`gyro_noise_density`
Gyroscope bias drift	RRW / bias random walk	`gyroscope_random_walk`	`gyro_bias_random_walk`
Gyroscope bias stability	Bias instability	-	`gyro_bias_instability`

Per-sensor characterized values are documented on the individual IMU pages.

ON THIS PAGE

IMUView as Markdown