Camera frame moves crazy in RVIZ with D435i

[mhaboali]

Thanks for sharing this great project!

When I tried it with RealSense D435i using your docker method and your d435i.yaml config file, it worked but while moving it very little movement I saw it moved crazy in RVIZ and didn't stop at all as shown in the attached

Note: I didn't change anything except for topics names and paths

[Shame-fight]

As far as I know, many people have the same problem, including me. This phenomenon is very strange. Different cameras and different devices have encountered the same problem. I suspect that it is an algorithm problem or initialization. Troubled for a while, and do not know the solution. If you know some reasons or solve this problem, please let me know.

[xdtl]

Same here. It works great on MH_01_easy.bag. But when I use my own data, I got crazy camera trajectories. And every time the result is different with exactly the same setting. Could someone please give suggestions about how to identify and correct the problem? Thanks!

[jagauthier]

I'm also having this issue from a rosbag file I created with an realsense R200.

[BarzelS]

Any updates? I'm having the same problem?

[hanzheteng]

Perhaps you need to calibrate your IMU first

[BarzelS]

Perhaps you need to calibrate your IMU first

I did it using the rs-imu-calibration tool by realsense. I also updated the IMU related parameters using Kalibr tool.
@hanzheteng have you managed to get a stable VIO with the D435i and VINS-FUSION? Have you tried it with the new realsense versions?
RealSense ROS v2.2.15
LibRealSense v2.36.0

[dssdyx]

I use realsense d455 to run VINS Fusion and get great result. I think d435i and d455 have similar config. There are some important points deciding whether the system is stable or not.

1. Extrinsic between imu and cameras. I think stable vins system must depend on good extrinsic and we can use kalibr to solve it. The result of kalibr is Tci , but in vins config, body_T_cam seems to mean Tic.
2. Imu parameters. I used Allan variance to estimate imu noise and random walk. Actually, it's the main reason why I always get crazy result. Using default imu parameters such as 0.1, 0.001 will get good result. I think it's because when I calibrated imu, the system was static without vibration,which was different in moving situation. Imu noise and random walk should be increased in fact.
3. Synchronization problem. This is a little chaotic problem, because different realsense versions and firmwares have different situation.I now use librealsense 2.42, realsense-ros 2.2.22, which is ok. Also, remember calibrate time offset between cameras and imu in Kalibr using "--time-calibration" flag

[Camilochiang]

I used a RS415. I believe that the most important point, as mentioned by @dssdyx is the sync. After calibration with Klibr I got a value of 0.3
Here with td:0 (it fail)

Here with td:0.3