addFrame by combining two existing frames

[arjung128]

Hello, I would like to create a new frame based on two existing frames. Namely, joint_gripper_fingertip_right and joint_gripper_fingertip_left exist, and I would like to keep track of the center point between these two frames, i.e. the average position of the fingertips.

I know the following is the syntax:

frame = pin.Frame(
    name="joint_gripper_fingertip_center",
    parent_joint=2,
    parent_frame=6,
    placement=pin.SE3(
        rotation=np.eye(3),
        translation=np.array([0, 0, 0]),
    ),
    type=pin.FrameType.BODY,
)
model.addFrame(frame)

What I am unsure about is what the parent_joint and parent_frame in this situation would be. Also, how can the placement be made to be the average of the positions of the left and right gripper fingertips instead of a hardcoded pose? And pin.FrameType.BODY would be fine as the type for this, correct?

I'm sorry if this is a silly question. I tried looking through the other issues as well as the examples but couldn't quite find anything to help me beyond this point. Any help will be much appreciated, thanks in advance!

[jcarpent]

What I am unsure about is what the parent_joint and parent_frame in this situation would be.

parent_frame attribute is there just for the record of the underlying tree of frames.
parent_joint is the quantity you should take care

And pin.FrameType.BODY would be fine as the type for this, correct?

You should rather use OP_FRAME which stands for Operational Frame, which seems to be aligned with your purposes as you are considering a virtual, non-physical, point.

Also, how can the placement be made to be the average of the positions of the left and right gripper fingertips instead of a hardcoded pose?

The fact that your two joints are moving independently (joint_gripper_fingertip_right and joint_gripper_fingertip_left), you should consider updating the placement quantity of your new frame after each change of configuration, as your averaged placement will change according to the relative placement of the two considered frames.

[arjung128]

Thanks so much for your prompt response!

parent_joint is the quantity you should take care

Would the parent_joint be the same as that of joint_gripper_fingertip_right? Here is the description for joint_gripper_fingertip_right from the URDF:

<joint name="joint_gripper_fingertip_right" type="fixed">
  <origin xyz="-0.190596948563868 -0.015 0.0" rpy="-1.57079632679483 -3.43320051448326e-14 0.540456056432235"/>
  <axis xyz="0.0 0.0 0.0"/>
  <parent link="link_gripper_finger_right"/>
  <child link="link_gripper_fingertip_right"/>
</joint>

Would I use this information to set parent_joint? Or what should I set the parent_joint to, and where can I get this information? Apologies, I'm still unsure about what this quantity should be set to.

Once again, thanks a lot for your work and for your help!

[jcarpent]

Yes, you can use as parent joint one of the two joints already identified.

[arjung128]

Got it, thanks!

Here is my code:

model, geom_model, visual_model = pin.buildModelsFromUrdf('stretch/stretch-re2-2025/exported_urdf/stretch.urdf', 'stretch/stretch-re2-2025/exported_urdf/')

frame = pin.Frame(
    name="joint_gripper_fingertip_center",
    parent_joint=11, # model.parents[-2] = model.parents[-3] = 11 (parents of the right and left fingertip joints)
    parent_frame=81, # model.frames[81].name = 'joint_gripper_fingertip_right'
    placement=pin.SE3(
        rotation=np.eye(3),
        translation=np.array([0, 0, 0]),
    ),
    type=pin.FrameType.OP_FRAME,
)
model.addFrame(frame)

data, geom_data, visual_data = pin.createDatas(model, geom_model, visual_model)

q = pin.randomConfiguration(model)
pin.forwardKinematics(model, data, q)
pin.framesForwardKinematics(model, data, q)

# set the translation for joint_gripper_fingertip_center to be the average of joint_gripper_fingertip_right and joint_gripper_fingertip_left
data.oMf[model.getFrameId('joint_gripper_fingertip_center')].translation = (data.oMf[model.getFrameId('joint_gripper_fingertip_right')].translation + data.oMf[model.getFrameId('joint_gripper_fingertip_left')].translation) / 2

However, this doesn't seem to affect pin.computeFrameJacobian(model,data,q,model.getFrameId('joint_gripper_fingertip_center')), and calling pin.updateFramePlacements(model, data) reverts the joint_gripper_fingertip_center frame to a completely different pose. Any ideas what I'm missing to get an accurate Jacobian (using pin.computeFrameJacobian) and frame placement (using pin.updateFramePlacements) after manually setting the translation to be the midpoint between the two finger joints?