In this project, the grasp synthesis network is based on KPConv, and we use GraspIt! to view the results. Please configure the development environment according to the instructions in links.
Installation process:
-
Download this code and unzip it. Note: 'cpp_wrappers', etc. are from [KPConv].
-
KPConv: It is recommended to install the ubuntu version.
-
GraspIt!: It is recommended to install the ubuntu version.
-
If your research continues to be based on the above two projects, please directly cite the original work.
-
In order to facilitate understanding, this code is based on the Barretthand with low degrees of freedom (the model and kinematics calculation are in './utils'). we also omit the pre-training process, etc., which require additional data generation and processing. If you are interested, you can add it by referring to the paper or contact us.
-
Unzip './datasets/functional_area(0-2).zip' to the './datasets/functional_area/'.
-
Run the following code to automatically process data and start network training :
python train_GraspNet.py -
The results are in the folder './results', which can be viewed by directly using [GraspIt!], or copy all .xml files that you want to view to the 'your-graspit-root-dir/worlds/', and then run the following code :
python show_result.py
- Our initial research has been accepted by ICCV2021.
- We have completed the real robot experiment, and the results will be published together with the new paper.
If you find our work useful in your research, please consider citing:
@InProceedings{Zhu_2021_ICCV,
author = {Zhu, Tianqiang and Wu, Rina and Lin, Xiangbo and Sun, Yi},
title = {Toward Human-Like Grasp: Dexterous Grasping via Semantic Representation of Object-Hand},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month = {October},
year = {2021},
pages = {15741-15751}
}