Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/ethz-asl/kalibr/llms.txt

Use this file to discover all available pages before exploring further.

The imu.yaml file provides Kalibr with the stochastic noise characteristics of your IMU. These parameters are used as prior information during the continuous-time batch optimization in kalibr_calibrate_imu_camera. Incorrect values will bias the calibration results, so they should come from the IMU datasheet or be identified through Allan variance analysis.

​
Example

The following example is taken directly from the Kalibr source for an ADIS16448 IMU: 
accelerometer_noise_density: 0.006   # m/s^2/sqrt(Hz)
accelerometer_random_walk:   0.0002  # m/s^3/sqrt(Hz)
gyroscope_noise_density:     0.0004  # rad/s/sqrt(Hz)
gyroscope_random_walk:       4.0e-06 # rad/s^2/sqrt(Hz)
update_rate:                 200.0   # Hz
rostopic:                    /imu0

​
Field reference

​
accelerometer_noise_density
number
required
Continuous-time noise spectral density of the accelerometer, in m/s^2/sqrt(Hz). This is also called the velocity random walk. Corresponds to the white noise floor of the accelerometer Allan variance curve. Must be positive.
​
accelerometer_random_walk
number
required
Accelerometer bias instability (bias random walk), in m/s^3/sqrt(Hz). Corresponds to the rate random walk region of the accelerometer Allan variance curve. Must be positive.
​
gyroscope_noise_density
number
required
Continuous-time noise spectral density of the gyroscope, in rad/s/sqrt(Hz). This is also called the angle random walk. Corresponds to the white noise floor of the gyroscope Allan variance curve. Must be positive.
​
gyroscope_random_walk
number
required
Gyroscope bias instability (bias random walk), in rad/s^2/sqrt(Hz). Corresponds to the rate random walk region of the gyroscope Allan variance curve. Must be positive.
​
update_rate
number
required
IMU measurement rate in Hz. Kalibr uses this to convert continuous-time noise densities to discrete-time noise covariances internally. Must be positive. Use the actual hardware rate, not a downsampled rate.
​
rostopic
string
required
ROS topic that publishes sensor_msgs/Imu messages in the bag file (e.g. /imu0).

​
Determining parameter values

From the datasheet. Most IMU manufacturers provide the angle random walk (ARW) and velocity random walk (VRW) in the datasheet. These are the noise densities. Bias instability figures in the datasheet correspond to the random walk parameters. From Allan variance analysis. For better accuracy, record a static IMU dataset and compute the Allan deviation curve. The slope of -½ region gives the noise density; the slope of +½ region gives the random walk. Tools such as allan_variance_ros automate this process.
Using values that are too small (overly optimistic) will cause the optimizer to over-trust IMU integration and can produce biased extrinsic results. When in doubt, use values slightly larger than the datasheet specifies.

​
Multiple IMUs

When calibrating multiple IMUs, pass one YAML file per IMU to --imu: 
kalibr_calibrate_imu_camera \
  --bag recording.bag \
  --cams camchain.yaml \
  --imu imu0.yaml imu1.yaml \
  --imu-models calibrated calibrated \
  --target aprilgrid.yaml
Each file has the same format. The first IMU listed is treated as the reference IMU for the continuous-time spline. See the multi-IMU calibration guide for details.

​
Relationship to output imu.yaml

kalibr_calibrate_imu_camera writes a <bagtag>-imu.yaml output file that wraps the calibrated IMU parameters under named sections (imu0, imu1, …). This output format differs from the input format: the input file contains parameters for a single IMU at the top level, while the output file groups multiple IMUs by name. See Output files for more details on the naming convention.

Build docs developers (and LLMs) love

Get started for freeTalk to us