Trajectory error

[DennisAdenHollander]

I am running the tunnel data from the Kimera-Multi Dataset using robots: sparkal1 and sparkal2 on the demo.py script. However the ground truth and the estimated trajectory have a large error, due to the use of wheel odometry data instead of the VIO odometry data. This VIO data is not provided by Kimera-Multi but is requested for roman/params/kimera_multi_data. Is using Kimera-VIO sufficient to decrease the trajectory error?

[mbpeterson70]

Hi @DennisAdenHollander, Kimera-VIO is definitely much better than the wheel odometry in the Kimera-Multi dataset. I haven't run ROMAN before with just the wheel odometry, but if the drift is significant enough, it could negatively impact the loop closure quality.

If you switch to wheel odometry, you want to also be certain that you set up the parameters right for setting the transforms between the base_link and camera. For Kimera-VIO, the odometry base_link is the same as the camera frame, so if you don't change this, the frames will all be out of sorts.

You would need to switch the pose data params from something like

pose_data:
  type: bag
  path: ${KMD}/kimera_vio/${ROBOT}.bag
  topic: /${ROBOT}/kimera_vio_ros/odometry
  time_tol: 0.5
  T_camera_flu:
    input_type: "string"
    string: "T_RDFFLU"

to something like

pose_data:
  type: bag
  path: ${KMD}/bags/${ROBOT}.bag
  topic: /${ROBOT}/jackal_velocity_controller/odom
  time_tol: 0.5
  T_camera_flu:
    input_type: "string"
    string: "T_RDFFLU"
  T_odombase_camera
    input_type: "string"
    string: "T_FLURDF"

although you are also introducing an error here since the cameras were above and slightly forward the center of the robots. I'm not sure those measurements were recorded.