Faild to load Leap Hand model

[tianhaowuhz]

Faild to load Leap Hand model

[tianhaowuhz]

It seems the problem is caused by some value in urdf is "e-40", which exceed the limit, after fix this, the model can be loaded.

[tianhaowuhz]

Although I have been able to load the model, the retarget result from human video in example seems a little weird, for example, at the initial frame the finger of human hand is open,

but the retargeted leap hand's finger is bent,

is there any parameter I should try to adjust this?
Here is my current yml config file of leap hand

  retargeting:
  type: vector
  urdf_path: leap_hand/leap_hand_right.urdf
  wrist_link_name: "base"

  # Target refers to the retargeting target, which is the robot hand
  target_joint_names: null
  target_origin_link_names: [ "base", "base", "base", "base" ]
  target_task_link_names: [ "thumb_fingertip", "fingertip", "fingertip_2", "fingertip_3" ]
  scaling_factor: 1.6

  # Source refers to the retargeting input, which usually corresponds to the human hand
  # The joint indices of human hand joint which corresponds to each link in the target_link_names
  target_link_human_indices: [ [ 0, 0, 0, 0 ], [ 4, 8, 12, 16 ] ]

  # A smaller alpha means stronger filtering, i.e. more smooth but also larger latency
  low_pass_alpha: 0.2

[yzqin]

The yml configuration file seems to be in order; there don't appear to be any faults with it. The actual issue lies in the positioning of the fingertip link—it isn't accurately positioned at the fingertip of the leap hand as one might expect. The frame definition for the fingertip link is presented below:

[yzqin]

The solution is to add a finger tip virtual link for leap hand so that the retargeting computation can be performed easily.

Please check the recent commit 3aa3e1a. We provide a yml config file for the leap hand for both vector retargeting and dexpilot retargeting. You can use our provided model in this repo.

[yzqin]

Here is the video generated based on DexPilot config of leap hand

leap.mp4

[tianhaowuhz]

It do looks better, thank you!