dexsuite / dex-urdf

This repository offers an assortment of high-quality models for dexterous hands and objects. Both of them are in URDF format.

Robot Model	Visual1	Collision2
Allegro Hand
Shadow Hand
SCHUNK SVH Hand
Ability Hand
Leap Hand
DClaw Gripper
Barrett Hand
Inspire Hand
Panda Gripper

URDF Parser Links: yourdfpy, IsaacGym, SAPIEN, PyBullet

Robot Source

Robot Model	Official Website	URDF Source	CAD Model Source	License
Allegro Hand	Wonik Robotics	allegro_hand_ros	N/A	BSD
Shadow Hand	Shadow Robot Company	sr_common	N/A	GPL-3.0
SCHUNK SVH Hand	SCHUNK	schunk_svh_ros_driver	N/A	Apache-2.0
Ability Hand	PSYONIC	ability-hand-api	N/A	N/A
Leap Hand	Leap Hand	LEAP_Hand_Sim	Leap Hand CAD	CC BY-NC 4.0
DClaw Gripper	Robel Benchmark	N/A	D'Claw CAD	Apache-2.0
Barrett Hand	Barrett Technology	bhand_model	BarrettHand CAD	BSD
DexHand	DexHand	dexhand_description	Dexhand	CC BY-NC-SA 4.0
Inspire Hand	Inspire-Robot	N/A	inspire_hand	CC BY-NC-SA 4.0

Citation

If you use dex_urdf in your work, please use the following citation:

@article{bunny-visionpro,
    title={Bunny-VisionPro: Real-Time Bimanual Dexterous Teleoperation for Imitation Learning},
    author={Runyu Ding and Yuzhe Qin and Jiyue Zhu and Chengzhe Jia and Shiqi Yang and Ruihan Yang and Xiaojuan Qi and Xiaolong Wang},
    year={2024},
    url={https://arxiv.org/abs/2407.03162},
}

This repository contains models derived from open-source contributions by various developers and manufactures. The development of dex_urdf was made possible thanks to the generous work of numerous designers and engineers. We extend our sincere appreciation to all those whose efforts have contributed to this project.

Footnotes

1. Ray tracing animation are rendered in SAPIEN using the urdf with glb version. Code can be found in generate_urdf_animation_sapien.py. ↩

2. Collision mesh are rendered in SAPIEN using the same urdf as the visual. Blue links are modeled using primitives while green links are modeled using convex triangle meshes. Code can be found in generate_urdf_collision_figure_sapien.py](tools/generate_urdf_collision_figure_sapien.py). ↩

dex-urdf's Issues

About inspire hand support

I'm excited to see that you updated the inspire hand commit yesterday ! Is it available on dex-retargeting now ?
Thanks for your amazing job :)

Which version of sapien is used?

error info

File "generate_sapien_collision_mesh.py", line 100, in <module>
   tyro.cli(main)
 File "/home/zhanghaoyang/.conda/envs/sapien/lib/python3.8/site-packages/tyro/_cli.py", line 184, in cli
   output = _cli_impl(
 File "/home/zhanghaoyang/.conda/envs/sapien/lib/python3.8/site-packages/tyro/_cli.py", line 446, in _cli_impl
   out, consumed_keywords = _calling.call_from_args(
 File "/home/zhanghaoyang/.conda/envs/sapien/lib/python3.8/site-packages/tyro/_calling.py", line 241, in call_from_args
   return unwrapped_f(*positional_args, **kwargs), consumed_keywords  # type: ignore
 File "generate_sapien_collision_mesh.py", line 93, in main
   parse_urdf(
 File "generate_sapien_collision_mesh.py", line 64, in parse_urdf
   if isinstance(collision_shape, sapien.physx.PhysxCollisionShapeConvexMesh):
AttributeError: module 'sapien.core' has no attribute 'physx'

It seems the version of sapien is not lateset. Could you help me with this? Thank you.

Could you provide the UR10e assets corresponding to Sapien?

Hi, @yzqin,

Could you provide the UR10e assets corresponding to Sapien? When I load the UR10e assets from Isaacgym, they blow up but can run well in Isaacgym.

Could you help me solve the problem? Thanks for any questions :)

troublesome leap_hand_left.urdf

Hi! About the left side of leap hand urdf, did you notice the collision of the mesh seems not that correct? This means, the thumb of leap hand left can not rotate freely and can easily get stuck with palm link. Is there any new version of leap hand urdf left?

Training speed is very slow by using shadow_hand_right.urdf

Hi, @yzqin

I try to use the [shadow_hand_right.urdf](https://github.com/dexsuite/dex-urdf/blob/main/robots/hands/shadow_hand/shadow_hand_right.urdf) with ur10e.urdf in Sapine Simulator for dexterous manipulation. However, I found when I use shadow_hand_right.urdf with ur10e.urdf as my robot, the training speed is very low! This is very strange!

But when I use xarm6_allegro_wrist_mounted_rotate.urdf in Project [dexart-release](https://github.com/Kami-code/dexart-release) as my robot, the training speed is very fast!

I didn't change any RL setting, but it is very slow caused by replacing the robot :(

Since the urdf files and meshes are larger than 25MB, I will send you ShadowHand_with_UR10e.zip and corresponding setting env/rl_env/base.py and utils/common_robot_utils.py like [dexart-release](https://github.com/Kami-code/dexart-release) and dex-hand-teleop by Email.

Thanks for any help!

No Collision in Sapien

Hi! Great work of yours!

I'm trying to use Sapien as my simulator. However, when I ran the examples, I noticed there is no physical collision in the hand, as shown below. Is this a bug, or is it a common occurrence in Sapien?

Request on updateing the support for the second-generation inspire hand

Hi, @yzqin
Thanks for this great work!

I found the project has supported the first-generation Inspire Hand :)
Could you update the support for the second-generation Inspire Hand by considering the Project?

Thanks for any help!

Best,
Charlie

Can you provide left shadow-hand robot files please.

How to attach hand on robot arm

Hi! Thanks for your exciting work. I want to know how to attach hand on robot arm. Is it just a matter of stitching together the URDF files? Can you provide some tutorials?