IMU

Supported on:RVC2 RVC4

IMU (inertial measurement unit) node can be used to receive data from the IMU chip on the device.Luxonis devices use different IMU stacks depending on platform generation:

RVC2 devices use either BNO08X or BMI270
RVC4 devices use LSM6DSV together with an AK09919 magnetometer.

The IMU chip is connected to the RVC over SPI. See OAK Hardware documentation to check whether your OAK camera has IMU integrated.

Looking for the conceptual docs?

For sensor families, output semantics, reference frames, factory calibration, and noise characterization, see the IMU hardware reference. For practical usage, see the IMU examples below.

How to place it

Python

1pipeline = dai.Pipeline()
2imu = pipeline.create(dai.node.IMU)

C++

1dai::Pipeline pipeline;
2auto imu = pipeline.create<dai::node::IMU>();

Inputs and Outputs

Pipeline flow

The usual IMU node flow is:

Create dai::node::IMU
Enable one or more IMUSensor reports at a requested rate
Configure batching with setBatchReportThreshold() and setMaxBatchReports()
Consume IMUData packets from the host output queue

Limitations

On RVC2 devices with BNO08X, gyroscope frequencies above 400 Hz can produce occasional jitter due to sensor hardware limitations.

Requested report rate

imu.enableIMUSensor(..., reportRate) accepts a requested report rate in Hz.Actual runtime behavior depends on the IMU and platform:

On RVC2 devices using BNO08X, requested rates round up to the next supported rate.
On RVC2 devices using BMI270, requested rates round down to the next supported rate. Requests above 400 Hz currently top out around 250 Hz.
On RVC4 devices using LSM6DSV + AK09919, treat validated runtime rates as platform-specific rather than assuming the silicon max ODR maps directly to the host stream rate.

For detailed per-sensor runtime behavior and hardware characteristics, see the IMU hardware reference.

High-rate raw streaming used in examples

The current depthai-core IMU examples and tests use the following raw-stream rates for high-rate capture:

ACCELEROMETER_RAW at 480 Hz
GYROSCOPE_RAW at 400 Hz

Batching

The IMU node batches packets before sending them to the host.

setBatchReportThreshold(N) sets the minimum number of IMU packets that should be ready before the device is allowed to send a batch
setMaxBatchReports(M) sets the maximum number of packets included in one batch

Higher batching can reduce host overhead and help when USB bandwidth or host scheduling becomes the bottleneck.

Packet fields and report semantics

Each IMUData message contains one or more IMUPacket entries. Each packet only carries the fields for the reports that were enabled in the node configuration.Typical fields include:

acceleroMeter
gyroscope
magneticField
rotationVector

Use the report families like this:

*_RAW: direct sensor output in the sensor-native frame
*_UNCALIBRATED: rotated into the Luxonis RDF frame using imuExtrinsics, without IMU calibration parameters applied
*_CALIBRATED: rotated into the Luxonis RDF frame and corrected using IMU calibration parameters
fused outputs such as ROTATION_VECTOR and GAME_ROTATION_VECTOR: forwarded from the sensor's internal processing path

Usage

Python

1pipeline = dai.Pipeline()
2imu = pipeline.create(dai.node.IMU)
3
4# enable ACCELEROMETER_RAW and GYROSCOPE_RAW at 100 hz rate
5imu.enableIMUSensor([dai.IMUSensor.ACCELEROMETER_RAW, dai.IMUSensor.GYROSCOPE_RAW], 100)
6# above this threshold packets will be sent in batch of X, if the host is not blocked and USB bandwidth is available
7imu.setBatchReportThreshold(1)
8# maximum number of IMU packets in a batch, if it's reached device will block sending until host can receive it
9# if lower or equal to batchReportThreshold then the sending is always blocking on device
10# useful to reduce device's CPU load  and number of lost packets, if CPU load is high on device side due to multiple nodes
11imu.setMaxBatchReports(10)

C++

1dai::Pipeline pipeline;
2auto imu = pipeline.create<dai::node::IMU>();
3
4// enable ACCELEROMETER_RAW and GYROSCOPE_RAW at 100 hz rate
5imu->enableIMUSensor({dai::IMUSensor::ACCELEROMETER_RAW, dai::IMUSensor::GYROSCOPE_RAW}, 100);
6// above this threshold packets will be sent in batch of X, if the host is not blocked and USB bandwidth is available
7imu->setBatchReportThreshold(1);
8// maximum number of IMU packets in a batch, if it's reached device will block sending until host can receive it
9// if lower or equal to batchReportThreshold then the sending is always blocking on device
10// useful to reduce device's CPU load  and number of lost packets, if CPU load is high on device side due to multiple nodes
11imu->setMaxBatchReports(10);

Report availability by hardware

Current public API scope

The current depthai-core IMU enum and docstrings are still BNO08X-oriented. In the bindings, the node is still documented as IMU node for BNO08X, so fused and processed report coverage should be read together with the hardware docs and validated on the target device when the sensor path is not BNO08X.

Report family	`IMUSensor` values	BNO08X	BMI270	LSM6DSV	AK09919	Notes
Raw acceleration	`ACCELEROMETER_RAW`	Yes	Yes	Yes	No	Raw accelerometer stream
Accelerometer in Luxonis frame, without IMU calibration parameters	`ACCELEROMETER_UNCALIBRATED`	Yes	Yes	Yes	No	`RAW -> UNCALIBRATED` applies `imuExtrinsics`. This is the unified IMU-frame accelerometer path.
Accelerometer in Luxonis frame, with IMU calibration parameters	`ACCELEROMETER_CALIBRATED`	Yes	Yes	Yes	No	Requires IMU calibration parameters to be present in the runtime calibration payload.
Raw angular velocity	`GYROSCOPE_RAW`	Yes	Yes	Yes	No	Raw gyroscope stream
Gyroscope in Luxonis frame, without IMU calibration parameters	`GYROSCOPE_UNCALIBRATED`	Yes	Yes	Yes	No	`RAW -> UNCALIBRATED` applies `imuExtrinsics`. On older devices this is expected to work out of the box.
Gyroscope in Luxonis frame, with IMU calibration parameters	`GYROSCOPE_CALIBRATED`	Yes	Yes	Yes	No	Requires IMU calibration parameters to be present in the runtime calibration payload.
Raw magnetic field	`MAGNETOMETER_RAW`	Yes	No	No	Yes	On RVC4, raw magnetometer data comes from the AK09919 companion sensor.
Processed magnetic field	`MAGNETOMETER_CALIBRATED`, `MAGNETOMETER_UNCALIBRATED`	Yes	No	No	Validate on target hardware	Public enum/docstrings are still BNO08X-shaped for processed magnetometer outputs.
Derived acceleration outputs	`LINEAR_ACCELERATION`, `GRAVITY`	Yes	No	Validate on target hardware	No	BNO08X exposes these directly. For non-BNO paths, validate availability on the target device and branch.
Fused orientation outputs	`ROTATION_VECTOR`, `GAME_ROTATION_VECTOR`, `GEOMAGNETIC_ROTATION_VECTOR`, `ARVR_STABILIZED_ROTATION_VECTOR`, `ARVR_STABILIZED_GAME_ROTATION_VECTOR`	Yes	No	Validate on target hardware	No standalone	These are sensor-fusion outputs, not host-side fusion done by the IMU node.

Calibration and extrinsics

Use device.readCalibration() to read the factory calibration payload from the device. Use device.getCalibration() and device.setCalibration() when working with runtime calibration overrides.For the processed inertial outputs, use this model:

RAW -> UNCALIBRATED: align the sensor-native frame into the Luxonis RDF frame using imuExtrinsics
UNCALIBRATED -> CALIBRATED: apply the accelerometer or gyroscope calibration set in the IMU calibration parameters
ACCELEROMETER_UNCALIBRATED and GYROSCOPE_UNCALIBRATED therefore share the unified IMU-frame path
ACCELEROMETER_CALIBRATED and GYROSCOPE_CALIBRATED add the runtime calibration correction on top of that

Processed output transforms

RAW -> UNCALIBRATED

Python

1import depthai as dai
2import numpy as np
3
4with dai.Pipeline() as pipeline:
5    imu = pipeline.create(dai.node.IMU)
6    imu.enableIMUSensor(dai.IMUSensor.ACCELEROMETER_RAW, 100)
7    imu_q = imu.out.createOutputQueue(maxSize=10, blocking=False)
8
9    device = pipeline.getDefaultDevice()
10    calib = device.readCalibration()
11    imu_to_cam = np.array(calib.getImuToCameraExtrinsics(dai.CameraBoardSocket.CAM_A, False))
12    R_imu_to_cam = imu_to_cam[:3, :3]
13
14    pipeline.start()
15    pkt = imu_q.get().packets[0]
16    raw = np.array([pkt.acceleroMeter.x, pkt.acceleroMeter.y, pkt.acceleroMeter.z])
17    uncalibrated = R_imu_to_cam @ raw

UNCALIBRATED -> CALIBRATED

Python

1import depthai as dai
2import numpy as np
3
4imu_calibration = [
5    [1.0, 0.0, 0.0, 0.125],
6    [0.0, 1.0, 0.0, 0.000],
7    [0.0, 0.0, 1.0, 0.000],
8]
9
10with dai.Pipeline() as pipeline:
11    imu = pipeline.create(dai.node.IMU)
12    imu.enableIMUSensor(dai.IMUSensor.ACCELEROMETER_UNCALIBRATED, 100)
13    imu_q = imu.out.createOutputQueue(maxSize=10, blocking=False)
14
15    pipeline.start()
16    pkt = imu_q.get().packets[0]
17    uncalibrated = np.array([pkt.acceleroMeter.x, pkt.acceleroMeter.y, pkt.acceleroMeter.z])
18
19    calibration = np.array(imu_calibration)
20    calibrated = calibration[:, :3] @ uncalibrated + calibration[:, 3]

Reading multiple IMU streams

If the target hardware exposes the selected reports, you can enable multiple IMU report families and read only the fields that were requested.

Python

1import depthai as dai
2
3with dai.Pipeline() as pipeline:
4    imu = pipeline.create(dai.node.IMU)
5
6    imu.enableIMUSensor(dai.IMUSensor.ACCELEROMETER_RAW, 100)
7    imu.enableIMUSensor(dai.IMUSensor.GYROSCOPE_RAW, 100)
8    imu.enableIMUSensor(dai.IMUSensor.MAGNETOMETER_RAW, 100)
9
10    imu.setBatchReportThreshold(1)
11    imu.setMaxBatchReports(10)
12
13    imu_q = imu.out.createOutputQueue(maxSize=50, blocking=False)
14
15    pipeline.start()
16    while pipeline.isRunning():
17        imu_data = imu_q.get()
18        for pkt in imu_data.packets:
19            a = pkt.acceleroMeter
20            g = pkt.gyroscope
21            m = pkt.magneticField

C++

1#include "depthai/depthai.hpp"
2
3int main() {
4    dai::Pipeline pipeline;
5    auto imu = pipeline.create<dai::node::IMU>();
6
7    imu->enableIMUSensor(dai::IMUSensor::ACCELEROMETER_RAW, 100);
8    imu->enableIMUSensor(dai::IMUSensor::GYROSCOPE_RAW, 100);
9    imu->enableIMUSensor(dai::IMUSensor::MAGNETOMETER_RAW, 100);
10
11    imu->setBatchReportThreshold(1);
12    imu->setMaxBatchReports(10);
13
14    auto imuQ = imu->out.createOutputQueue(50, false);
15
16    pipeline.start();
17    while(pipeline.isRunning()) {
18        auto data = imuQ->get<dai::IMUData>();
19        for(const auto& pkt : data->packets) {
20            const auto& a = pkt.acceleroMeter;
21            const auto& g = pkt.gyroscope;
22            const auto& m = pkt.magneticField;
23        }
24    }
25}

Examples of functionality

Reference

class

dai::node::IMU

#include IMU.hpp

variable

Output out

Outputs IMUData message that carries IMU packets.

variable

Input mockIn

Mock IMU data for replaying recorded data

function

void enableIMUSensor(IMUSensorConfig sensorConfig)

Enable a new IMU sensor with explicit configuration

function

void enableIMUSensor(const std::vector< IMUSensorConfig > & sensorConfigs)

Enable a list of IMU sensors with explicit configuration

function

void enableIMUSensor(IMUSensor sensor, uint32_t reportRate)

Enable a new IMU sensor with default configuration

function

void enableIMUSensor(const std::vector< IMUSensor > & sensors, uint32_t reportRate)

Enable a list of IMU sensors with default configuration

function

void setBatchReportThreshold(std::int32_t batchReportThreshold)

Above this packet threshold data will be sent to host, if queue is not blocked

function

std::int32_t getBatchReportThreshold()

Above this packet threshold data will be sent to host, if queue is not blocked

function

void setMaxBatchReports(std::int32_t maxBatchReports)

Maximum number of IMU packets in a batch report

function

std::int32_t getMaxBatchReports()

Maximum number of IMU packets in a batch report

function

void enableFirmwareUpdate(bool enable)

Whether to perform firmware update or not. Default value: false.

inline function

DeviceNodeCRTP()

inline function

DeviceNodeCRTP(const std::shared_ptr< Device > & device)

inline function

DeviceNodeCRTP(std::unique_ptr< Properties > props)

inline function

DeviceNodeCRTP(std::unique_ptr< Properties > props, bool confMode)

inline function

DeviceNodeCRTP(const std::shared_ptr< Device > & device, std::unique_ptr< Properties > props, bool confMode)

Need assistance?

Head over to Discussion Forum for technical support or any other questions you might have.

ON THIS PAGE

IMUView as Markdown

How to place it

Python

C++

Inputs and Outputs

Pipeline flow

Limitations

Requested report rate

High-rate raw streaming used in examples

Batching

Packet fields and report semantics

Usage

Python

C++

Report availability by hardware

Calibration and extrinsics

Processed output transforms

RAW -> UNCALIBRATED

UNCALIBRATED -> CALIBRATED

Reading multiple IMU streams

Python

C++

Examples of functionality

Reference

dai::node::IMU

Need assistance?

IMU