Analysis on Potential Yaw Estimation Errors

[SKT-r]

Background
There is a phenomenon observed occasionally in RViz where the raw LiDAR point cloud be shifted in the yaw direction relative to the NDT point cloud (point cloud map). This suggests a possibility of incorrect yaw estimation.

The observed phenomenon is as follows:
No misalignment on Rviz

Misalignment on Rviz

Isolate the issue from RViz
Considering the aforementioned issue could be a rendering problem in RViz, I conducted experiments by creating a custom script. Here is an overview:

• Using sqlite3 and rosbags, I extracted the following topics from the bag file.

  • NDT point cloud
  • raw LiDAR point cloud
  • self-position

• I extracted the raw LiDAR point cloud at a specific timestamp and derived the most appropriate self-position using linear interpolation.

• Using the derived self-position, I transformed the raw LiDAR point cloud into the global coordinate system and overlaid it onto the point cloud map.

As a result, I obtained similar results to what was observed in RViz.

How to proceed
One possibility to consider is that the yaw_bias of the EKF (Extended Kalman Filter) may be having a detrimental effect. Investigating this would be a logical next step.

[chivas1000]

would it be the sensor kit calibration issue?
see my issue at #4788