ACRONYM: A Large-Scale Grasp Dataset Based on Simulation
Clemens Eppner, Arsalan Mousavian, Dieter Fox
We introduce ACRONYM, a dataset for robot grasp planning based on physics simulation. The dataset contains 17.7M parallel-jaw grasps, spanning 8872 objects from 262 different categories, each labeled with the grasp result obtained from a physics simulator. We show the value of this large and diverse dataset by using it to train two state-of-the-art learning-based grasp planning algorithms. Grasp performance improves significantly when compared to the original smaller dataset. This website contains links to download the dataset and tools to work with it in your own research and applications. Don't hesitate to contact us for further questions.
@inproceedings{acronym2020,
title = {{ACRONYM}: A Large-Scale Grasp Dataset Based on Simulation},
author = {Eppner, Clemens and Mousavian, Arsalan and Fox, Dieter},
year = {2020},
booktitle = {2021 {IEEE} Int. Conf. on Robotics and Automation, {ICRA}},
}
Python helper tools and two test files from the full dataset: https://github.com/NVlabs/acronym
Full Download (1.6GB): Link
The dataset is comprised of 8836 HDF5 files. File names follow the scheme <category>_<shapenetid>_<scale>.h5 .
Each file is structured as follows:
>> h5ls -r WineGlass_2d89d2b3b6749a9d99fbba385cc0d41d_0.0024652679182251653.h5
/ Group
/grasps Group
/grasps/qualities Group
/grasps/qualities/flex Group
/grasps/qualities/flex/object_in_gripper Dataset {2000}
/grasps/qualities/flex/object_motion_during_closing_angular Dataset {2000}
/grasps/qualities/flex/object_motion_during_closing_linear Dataset {2000}
/grasps/qualities/flex/object_motion_during_shaking_angular Dataset {2000}
/grasps/qualities/flex/object_motion_during_shaking_linear Dataset {2000}
/grasps/transforms Dataset {2000, 4, 4}
/gripper Group
/gripper/configuration Dataset {1}
/gripper/type Dataset {SCALAR}
/object Group
/object/com Dataset {3}
/object/density Dataset {SCALAR}
/object/file Dataset {SCALAR}
/object/friction Dataset {SCALAR}
/object/inertia Dataset {3, 3}
/object/mass Dataset {SCALAR}
/object/scale Dataset {SCALAR}
/object/volume Dataset {SCALAR}
HDF5 files can be read in Python using the h5py package. For example:
> import h5py
> import numpy as np
> grasps = h5py.File('WineGlass_2d89d2b3b6749a9d99fbba385cc0d41d_0.0024652679182251653.h5', 'r')
> grasps.visititems(print)
> grasps['object/file'][()]
'meshes/WineGlass/2d89d2b3b6749a9d99fbba385cc0d41d.obj'
> grasps['grasps/transforms'].shape
(2000, 4, 4)
> grasps['grasps/qualities/flex/object_in_gripper'].shape
(2000,)
> np.unique(grasps['grasps/qualities/flex/object_in_gripper'])
array([0, 1])
Under which license is the dataset published?
ACRONYM is licensed under CC BY-NC 4.0.
What is the reference frame for the Franka Panda Gripper model?
The grasp transforms describe the transformation of the Panda Gripper frame (attached to the gripper mesh as shown in the picture to the right) in the object's mesh coordinate system.
Goal-Auxiliary Actor-Critic for 6D Robotic Grasping with Point Clouds by L. Wang, Y. Xiang and D. Fox
Object Rearrangement Using Learned Implicit Collision Functions by M. Danielczuk, A. Mousavian, C. Eppner and D. Fox
Contact-GraspNet: Efficient 6-DoF Grasp Generation in Cluttered Scenes by M. Sundermeyer, A. Mousavian, R. Triebel and D. Fox
Reactive Human-to-Robot Handovers of Arbitrary Objects by W. Yang, C. Paxton, A. Mousavian, Y. Chao, M. Cakmak, D. Fox