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.

When a platform carries more than one IMU, you need to know the rigid-body transformation between them and any time offset in their timestamps. kalibr_calibrate_imu_camera supports multi-IMU setups by accepting multiple --imu YAML files. The first IMU listed becomes the reference sensor; all other IMUs are calibrated relative to it. At least one camera is required as an aiding sensor. The camera provides the visual observations that anchor the continuous-time trajectory, which in turn allows the spatial and temporal relationships between IMUs to be estimated.
Multi-IMU calibration is an extension of the camera-IMU workflow. Read the camera-IMU calibration guide first for background on the required inputs and the general calibration procedure.

​
How multi-IMU calibration works

The calibrator fits a single continuous-time B-spline trajectory to the visual observations from the camera chain. Each IMU is then registered against this common trajectory:
  • The reference IMU (first in the list) defines the body frame. Its pose relative to the camera chain is the primary calibration result, identical to the single-IMU case.
  • Additional IMUs have their orientation priors estimated by cross-correlating angular rate norms against the reference IMU, then refined jointly with all other parameters in the batch optimizer.
Time offsets between IMUs can be estimated from the same cross-correlation (--imu-delay-by-correlation) and are then held fixed or refined as design variables.

​
Prerequisites

  • A ROS bag with image topics from at least one camera and IMU topics from every IMU
  • camchain.yaml from a prior camera calibration run
  • One imu.yaml per IMU with noise parameters and the corresponding ROS topic
  • aprilgrid.yaml target configuration

​
Step-by-step calibration

1

Prepare one IMU YAML per sensor

Each IMU needs its own YAML file with the appropriate rostopic and noise parameters:
imu0.yaml (reference IMU)
rostopic: /imu0
update_rate: 200.0
accelerometer_noise_density: 0.006
accelerometer_random_walk:   0.0002
gyroscope_noise_density:     0.0004
gyroscope_random_walk:       4.0e-06
imu1.yaml (secondary IMU)
rostopic: /imu1
update_rate: 400.0
accelerometer_noise_density: 0.003
accelerometer_random_walk:   0.0001
gyroscope_noise_density:     0.0002
gyroscope_random_walk:       2.0e-06
The order of --imu arguments determines which sensor is the reference IMU. The first argument is always treated as the reference; its body-frame pose is fixed to the identity transformation.
2

Record a calibration bag

Motion requirements are the same as for single-IMU calibration: full 6-DOF excitation with the AprilGrid target visible in the camera throughout. Ensure that all IMU topics are being published at their expected rates.
If the IMUs have significantly different noise levels, longer recordings or more energetic motion will help the lower-quality IMU accumulate sufficient observations for a reliable calibration.
3

Run the calibrator with multiple IMUs

Pass one --imu and one --imu-models entry for each IMU in the same order:
kalibr_calibrate_imu_camera \
  --bag MYROSBAG.bag \
  --cam camchain.yaml \
  --imu imu0.yaml imu1.yaml \
  --imu-models calibrated calibrated \
  --target aprilgrid.yaml
The reference IMU is imu0.yaml (first). imu1.yaml will have its transformation relative to imu0 estimated by the optimizer.
4

Enable IMU-to-IMU time delay estimation

By default, no temporal calibration is performed between IMUs — only between the camera chain and the reference IMU. To also estimate the time delay between each secondary IMU and the reference IMU, add --imu-delay-by-correlation:
kalibr_calibrate_imu_camera \
  --bag MYROSBAG.bag \
  --cam camchain.yaml \
  --imu imu0.yaml imu1.yaml \
  --imu-models calibrated calibrated \
  --imu-delay-by-correlation \
  --target aprilgrid.yaml
The delay is estimated by cross-correlating the angular rate norm of the reference IMU against each secondary IMU and then refined during optimization.
5

Review the output

The output files follow the same naming convention as single-IMU calibration:
FileContents
MYROSBAG-camchain-imucam.yamlCamera chain with T_cam_imu for the reference IMU
MYROSBAG-imu.yamlParameters for all IMUs including T_i_b (transformation from body/reference IMU frame to each IMU) and time offsets
MYROSBAG-results-imucam.txtFull numeric results
MYROSBAG-report-imucam.pdfVisual report with residuals for all IMUs
In the results file, look for the T_i_b matrix for each non-reference IMU. This is the transformation from the body (reference IMU) frame to the secondary IMU frame. The time offset for each secondary IMU is also reported in seconds.

​
Using scale-misalignment models

You can mix IMU models across sensors. For example, to use the detailed scale-misalignment model for the reference IMU and calibrated for a secondary IMU: 
kalibr_calibrate_imu_camera \
  --bag MYROSBAG.bag \
  --cam camchain.yaml \
  --imu imu0.yaml imu1.yaml \
  --imu-models scale-misalignment calibrated \
  --target aprilgrid.yaml
The number of --imu-models entries must equal the number of --imu entries.

​
Troubleshooting

​
”The time ranges of the IMUs do not overlap”

The cross-correlation step requires the reference IMU and secondary IMU time series to overlap. If they do not, the tool exits with a fatal error. Check that both IMU topics are being recorded for the full duration of the bag.

​
Secondary IMU orientation prior fails

The orientation prior estimation for a secondary IMU uses angular rate cross-correlation. If the secondary IMU has very different noise characteristics or is nearly collinear with the reference IMU’s rotation axes, this step can fail. Try increasing the aggressiveness of the calibration motion to include more pronounced rotations.

Camera-IMU calibration

Prerequisite guide covering single-IMU calibration and IMU YAML format.

IMU noise model

How to determine noise density and random walk values for your IMU.

Build docs developers (and LLMs) love

Get started for freeTalk to us