imu¶

Where IMU measurements come from and how far they can be trusted. This also holds the transform to the LiDAR. Both that transform and the noise figures are measurements, and the calibration section of the documentation covers how to take them.

imu.topic¶

Topic carrying sensor_msgs/msg/Imu. Ignored while imu.driver.type selects an in-process driver. This has to be the raw device output. Calibrating against an orientation-filtered topic produces a wrong result.

Part of the starter configuration.

imu.qos.reliability¶

Anything other than Reliable drops messages under load and is not supported.

Accepted values: besteffort, reliable, systemdefault

Accepted values are case-insensitive.

Part of the starter configuration.

imu.initial_setup¶

How long to collect IMU measurements before the first scan is processed. These samples give the filter its initial gravity direction, so the platform has to stand still for at least this long after start-up.

Type: time.

imu.lidar_only¶

Ignore the IMU entirely and run the odometry on the Lidar alone, with the filter propagating at a constant twist between scans.

This exists to produce a trajectory for calibrating the IMU against, which is circular if the IMU helped produce it. It is not a navigation mode: with no IMU the gravity direction is unobservable, so roll and pitch are relative to wherever the sensor started and not to level, and there is nothing to carry the estimate through a stretch where the geometry is degenerate. The covariances below stop describing sensor noise and start describing how much the constant-twist model can be trusted, so raise them.

Type: boolean.

imu.ahrs_filtering.method¶

Orientation filtering Pelorus applies to incoming IMU measurements. Leave it at Disable when the driver or the device already filters, which is the usual case. The result is published on output.imu_ahrs_filtered_topic and is never fed back into the odometry.

Accepted values: createorientationonly, filteraccelerationandcreateorientation, filteraccelerationonly, disable

Accepted values are case-insensitive.

imu.gravity¶

Magnitude of gravity in m/s^2, used as the starting point for the filter. 9.81 is close enough everywhere on Earth. The filter refines it from there.

Type: number.

imu.covariance_gyroscope¶

How far the filter trusts the IMU between LiDAR corrections. Larger values mean less trust, so the scan match pulls the estimate further on each update.

The defaults are the FAST-LIO tuning values and are not a noise model of any particular sensor. dedrunk imu noise prints values measured from your own IMU.

Type: number.

imu.covariance_gyroscope_bias¶

Process noise on the gyroscope bias, i.e. how fast the filter lets that bias wander.

Type: number.

imu.covariance_acceleration¶

How far the filter trusts the accelerometer between LiDAR corrections.

Type: number.

imu.covariance_acceleration_bias¶

Process noise on the accelerometer bias.

Type: number.

imu.driver¶

imu.driver.type¶

Disabled keeps the existing external IMU input configured by imu.topic. Xsens and Ouster enable in-process drivers and require Pelorus to be built with their matching feature (or the aggregate sensor-drivers feature).

Accepted values: disabled, xsens, ouster

Accepted values are case-insensitive.

imu.driver.xsens¶

Serial connection to an Xsens MTi. Only used when type = Xsens.

imu.driver.xsens.port¶

Serial device path. A stable /dev/serial/by-id/... path is recommended over /dev/ttyUSB0 because USB enumeration order can change between boots.

imu.driver.xsens.baudrate¶

Serial baud rate in bits per second. Must be positive and must match the device.

Type: integer.

Minimum: 1 (inclusive).

Maximum: 4294967295 (inclusive).

imu.driver.xsens.frame_id¶

Frame assigned to native IMU measurements.

imu.driver.xsens.output_rate¶

Requested MTData2 output frequency in Hz when configure_device is true. Must be in 1..=2000 and supported by the connected Xsens model and baud rate.

Type: integer.

Minimum: 1 (inclusive).

Maximum: 2000 (inclusive).

imu.driver.xsens.coordinate_system¶

World-coordinate convention requested for orientation output: Acceleration and rate-of-turn remain expressed in the IMU body frame.

Accepted values: enu, ned, nwu

Accepted values are case-insensitive.

imu.driver.xsens.configure_device¶

If true, put the device in configuration mode, request UTC, calibrated acceleration, rate-of-turn, and quaternion MTData2 fields at output_rate, then return it to measurement mode. If false, the device must already emit all required fields in compatible precision and coordinates.

Type: boolean.

imu.driver.xsens.timestamp¶

imu.driver.xsens.timestamp.mode¶

Utc requires a present and valid MTData2 UTC field and rejects measurements whose clock differs from the host beyond max_host_skew. Host timestamps a complete decoded frame at receipt time and is useful before GNSS/UTC becomes valid, but includes serial transport and scheduling latency.

Accepted values: utc, host

Accepted values are case-insensitive.

imu.driver.xsens.timestamp.max_host_skew¶

Maximum accepted absolute difference between Xsens UTC and host UTC. Must be positive. Used by Utc mode. Host mode already uses host time.

Type: time.

Minimum: 0s (exclusive).

imu.driver.xsens.timestamp.startup_timeout¶

Maximum positive duration to wait for usable MTData2 input during startup. If Utc cannot be satisfied, Pelorus fails startup and says so.

Type: time.

Minimum: 0s (exclusive).

imu.from_lidar_transform¶

Where the IMU sits relative to the LiDAR. Measure this. A value taken from a drawing is often a few millimetres out, and a wrong lever arm shows up as drift that scales with how hard the platform turns. See the calibration section of the documentation.

imu.from_lidar_transform.use_tf_tree¶

Look the transform up in the TF tree instead of using the values below. The lookup happens once, when the first IMU message arrives.

Type: boolean.

imu.from_lidar_transform.translation¶

Transform from the LiDAR frame into the IMU frame, so a point in LiDAR coordinates becomes a point in IMU coordinates. dedrunk extrinsics lidar-imu prints both values in this form, ready to paste.

Type: array<length>.

imu.from_lidar_transform.rotation¶

Transform from the LiDAR frame into the IMU frame, so a point in LiDAR coordinates becomes a point in IMU coordinates. dedrunk extrinsics lidar-imu prints both values in this form, ready to paste.

Type: array<array<number>>.

imu.from_lidar_transform.estimate_online¶

Keep refining the extrinsic in the filter instead of holding it at the values above, which then serve as the starting point.

This is refinement, not calibration: the filter needs a starting guess close enough to converge, and the extrinsic is only observable while the platform is actually being excited. On a robot that mostly drives straight it will drift along the unobservable directions and get worse. Leave it off unless you know the motion excites all axes, and check the result against what you started with.

Type: boolean.

imu.calibration¶

Corrections applied to every IMU sample before the filter sees it.

imu.calibration.gyroscope¶

Measure the gyroscope bias at start-up and subtract it. The platform has to stand still while the first samples measurements arrive. Stored bias figures drift with temperature and between power cycles, so a value measured at start-up is more accurate than one read from a file.

Type: boolean.

imu.calibration.gyroscope.samples¶

How many measurements the start-up bias estimate averages over.

Type: integer.

Minimum: 0 (inclusive).

imu.calibration.parameter_file¶

Kalibr-format IMU calibration to apply to every sample: scale factors, axis misalignment and biases. Relative path "./" from this config file, or absolute "/".

Produced by https://codeberg.org/stelzo/dedrunk, or by Kalibr itself with --imu-models scale-misalignment.

imu.calibration.time_offset¶

Seconds the IMU clock runs ahead of the Lidar clock, as reported by dedrunk extrinsics lidar-imu. Pelorus subtracts it from every IMU timestamp, so paste the number exactly as dedrunk prints it, sign included.

This is a constant skew between the two sensors' clocks, not jitter. It matters most while turning: at 1 rad/s, 10 ms of skew smears the scan by about half a degree, which the registration then has to absorb.

Type: time.