Femto Bolt, URDF model wrong (?)

[fabioscap]

Hello,

I am suspecting that there is something wrong the Femto Bolt description. I am running an Ouster LiDAR and a Femto Bolt on the same robot. Here is my sensor setup:

Here is the URDF model:

By comparing the clouds from the LiDAR and the Bolt, I am expecting a small misalignment due to my limited precision in mounting the sensor. As a matter of fact there is (this is an angle between a wall and a ceiling):

So if I register the two scans I can get the actual offset between the LiDAR and Bolt:

But then the transform on the URDF looks way off and corrects way more than what is reasonably expected from human precision. There is no way the camera sits so far back:

I cannot really understand where my mistake could be. I made sure to disambiguate other sources of problems.

• done the LiDAR-Bolt extrinsic calibration at many different depth ranges, reliably obtaining the same offset.
• also, there is a robotic arm in front of the camera. You can see it in the white patch in the last screenshot. That patch also gets aligned (visually) to the actual arm when correcting the extrinsic. so I am tempted to exclude the LiDAR from the problem.
• the depth and color frames are aligned with each other as far as I can see plotting the colored point cloud.
• I am using branch v2-main
• I have tried undoing and doubling the tilt described in Coordinate System Issue with camera_color_frame and camera_color_optical_frame in Femto Bolt 1.5.7 #44, Coordinate System Orientation issue with camera_depth_optical_frame and camera_color_optical_frame in Femto Bolt #52 and related issues but that did not help. Not one single rotation can correct the cm-scale offset I have.

As a temporary solution, I have added a transform that corrects the visual model depth frame to the actual depth frame.
Here is the approximate transform:

  <link name="camera_link"/>
  <joint name="camera_visual_to_camera" type="fixed">
    <parent link="camera_visual_link"/>
    <child link="camera_link"/>
    <!-- <origin xyz="0.0 0.0 0.0" rpy="0.0 0.0 0.0"/> -->
    <origin xyz="-0.029750074843318527 -0.005784185950938374 -0.05113393129377081" rpy="-0.018286819223818494 0.017199897606273495 0.010042825389099981"/>
  </joint>

However, I would like to understand the root cause of this problem so it would be helpful if you could double check the URDF model or suggest me where the error might be. Thanks,
fabio

[christian-rauch]

Are you using the correct frames? I have attached the Femto Bolt, wrist-mounted, at the end of a robot arm. Using the forward kinematics of the arm, the CAD information from the mounting jig, and the TF from the camera model, I get a pretty good alignment of the point clouds when moving the arm around.

I would visualise the frames of your TF tree and see if they match with the real world. Then I would check if the colour and the registered depth image are given in the correct optical frame.

I noticed that some frames in the model are wrong. The IMU frame, for example, is outside the model: #103. But the frames for the cameras seem to be correct.

[fabioscap]

Hello, thanks for the reply!
I would like to think that I've used the correct frames, but I am going to triple check.
Also, I don't know if it's relevant, but I cannot reproduce #103. Here is what my tfs look like (I am using use_nominal_extrinsics="false" in the URDF and enable_publish_extrinsic: true in the launch file):

camera_link ->  camera_gyro_frame
- Translation: [-0.053, -0.038, 0.010]
- Rotation: in Quaternion [0.499, 0.498, -0.502, 0.501]

camera_link ->  camera_accel_frame
- Translation: [-0.053, -0.038, 0.010]
- Rotation: in Quaternion [0.499, 0.498, -0.502, 0.501]

[christian-rauch]

It looks like the issue I reported is only present with the v1 on the main branch and not v2 on the v2-main branch. See #103 (comment) for a direct comparison of the frame transforms between both versions.

[fabioscap]

Hello,

I just want to say that just this morning I found a bug in my URDF which is 90% likely to have caused my mismatch. I will correct it today but I wanted to let you know before other people put effort in this issue. Thanks a lot and sorry for the confusion!

EDIT: I was able to fix one of my bugs and corrected the vertical dispalcement, but not the horizontal one, as the camera is still inside the robot

[Danilrivero]

Hi, I imported the URDF file at orbbec_description back when I needed it, there are probably some errors as shown in #103 (comment). Open to any feedback I could check it back if required.

[ob-yalian]

@fabioscap We have updated the Femto Bolt URDF model in the latest v2-main branch (tag 2.5.5).
Please try the updated version, and let us know if you still see any misalignment or unexpected transforms.

Thanks again for your feedback!