rpng / open_vins

An open source platform for visual-inertial navigation research.
https://docs.openvins.com
GNU General Public License v3.0
2.06k stars 615 forks source link

Timeshift_cam_imu weird comportement #376

Closed Atohme closed 9 months ago

Atohme commented 10 months ago

Hi, I am working around openvins with a custom hardware around Ardupilot.

The behavior happen in the camera Imu configuration file.

My Kalibr calibration report me a time offset of -0.112xxxxx. When I put those value in the timeshift_cam_Imu parameter the ov_msckf crashed.

The error is removed by being under -0.1 of time offset,i.e, if I put -0.09 the ov_msckf doesn't crashed and the odometry is working perfectly.

Also, enabling the time offset online optimization of openvins report me the same value of -0.11xxx. So most likely something is wrong in this parameter.

I hope the dev can help me with that.

goldbattle commented 10 months ago

Please provide the configuration files along with a dataset which can reproduce this. Additionally, please post your calibration data and results from Kalibr (pdf and txt along with the command you have run).

On Tue, Aug 29, 2023 at 7:39 AM Atohme @.***> wrote:

Hi, I am working around openvins with a custom hardware around Ardupilot.

The behavior happen in the camera Imu configuration file.

My Kalibr calibration report me a time offset of -0.112xxxxx. When I put those value in the timeshift_cam_Imu parameter the ov_msckf crashed.

The error is removed by being under -0.1 of time offset,i.e, if I put -0.09 the ov_msckf doesn't crashed and the odometry is working perfectly.

Also, enabling the time offset online optimization of openvins report me the same value of -0.11xxx. So most likely something is wrong in this parameter.

I hope the dev can help me with that.

— Reply to this email directly, view it on GitHub https://github.com/rpng/open_vins/issues/376, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAQ6TYSQ2B6YFCXIOOZJYWTXXX5JNANCNFSM6AAAAAA4DCGQDU . You are receiving this because you are subscribed to this thread.Message ID: @.***>

goldbattle commented 9 months ago

Feel free to reopen if you have the dataset with the problem.