search

hyye / lio-mapping

Implementation of Tightly Coupled 3D Lidar Inertial Odometry and Mapping (LIO-mapping)
https://sites.google.com/view/lio-mapping
GNU General Public License v3.0
911 stars 321 forks source link

High default noise parameters #56

Closed lsdta closed 4 years ago

lsdta commented 4 years ago

I find that the IMU noise parameters affect odometry significantly, has anyone experimented with these values? (given that they are from VINS-mono). Lowering the values seems to give more weighting to the IMU for odometry, and vice-versa.

The default values of acc_n = 0.2 and gyr_n = 0.02 are quite high (a magnitude or two higher) compared to the noise values I found with this repo as per #39, is this done on purpose?

hyye commented 4 years ago

Is the repo you mentioned providing the continuous-time noise? IMU-Noise-Model can be another reference. Theoretically, these values should be obtained by calibration or empirically set for each sensor. The yaml file is an example of the configuration.

lsdta commented 4 years ago

I empirically calculated the noise values for my own sensor, and found that the acc_w and gyr_w values were of similar magnitude to your default values in the config files. However, the acc_n and gyr_n values were a magnitude or two lower in my tests vs your default values. The repo also provides their calculated parameters for the IMU you used here, which agree with my tests.

If you chose the high acc_n and gyr_n values in your config files through SLAM testing rather than noise measurements, then that's understandable, I am just looking to understand why as these are important parameters when it comes to tweaking outputs.