Question regarding ActionNet: How to extract 3D fingertip keypoints from dexterous hand DoF positions?
Hi team,
First of all, thank you for providing the ActionNet dataset! The data quality is great and it's been very helpful.
I have a technical question regarding the Fourier robot's dexterous hand data. Currently, the dataset provides the DoF positions (joint angles) for the hand, but my current pipeline requires the actual 3D coordinate keypoints for each fingertip.
I assumed I need to compute Forward Kinematics (FK) to get these coordinates, and I found the FFTAI/teleoperation repository. Specifically, I looked at robot_wrapper.py (Lines 269-289), which uses Pinocchio's framePlacement(q, frame_idx) to output the pin.SE3 pose (translation and rotation).
However, I'm not entirely sure if this is the correct implementation or URDF for the dexterous hand. For context, I am currently using the GR1T1_fourier_hand_6dof.urdf file and the gr1t1.yaml configuration.
When I compute and visualize the output keypoints using this specific setup, four of the fingers basically don't move, and only the thumb seems to be actuating. The visual result looks much more like a parallel gripper's behavior rather than a fully articulated dexterous hand.
My questions are:
Is this the correct script/method to get the 3D fingertip keypoints for the dexterous hand?
Given that I am using GR1T1_fourier_hand_6dof.urdf and gr1t1.yaml, is there a specific configuration or a different URDF I should be using to properly articulate all 5 fingers?
If this isn't the right approach, could you point me in the right direction to compute the fingertip FK?
Any guidance would be greatly appreciated. Thanks in advance!