This has some nice utility scripts and packages that allow for calculation of the noise values for use in both kalibr and IMU filters.
The dataset of the manufacture can find the "white noise" values for the system, but the bias noises need to be found through experimental tests.
The gyroscope_random_walk and accelerometer_random_walk values can normally be found on the IMU datasheet as either angular random walk or velocity random walk, respectively.
| Parameter | YAML element | Symbol | Units |
|---|---|---|---|
| Gyroscope "white noise" | gyroscope_noise_density |
||
| Accelerometer "white noise" | accelerometer_noise_density |
||
| Gyroscope "random walk" | gyroscope_random_walk |
||
| Accelerometer "random walk" | accelerometer_random_walk |
- With the IMU remaining still, record a ROS bag of the readings (we collected a bag for about 4 hours)
- Convert the ROS bag into a matlab mat file.
- Use the included
bagconvertROS package to do this - Example:
rosrun bagconvert bagconvert imu.bag /imu0
- Use the included
- Run the included matlab scripts to generate an allan deviation plot for the readings
- If using the parallel version, it uses the matlab parallel toolbox
- Need to specify the mat file that the bagconverter made, and the rate of IMU messages
- Interpret the generated charts to find noise values
- Run the process results script
- Will fit a -1/2 line to the left side of the allan plot
- White noise is at tau=1 (according to kalibr wiki)
- Will fit a 1/2 line to the right side of the allan plot
- Random walk is at tau=3 (according to kalibr wiki)
- Some example data can be found HERE:
- XSENS MTI-G-700
- Tango Yellowstone Tablet
- ASL-ETH VI-Sensor
