VL-Grasp

Official Implementation for paper "VL-Grasp: a 6-Dof Interactive Grasp Policy for Language-Oriented Objects in Cluttered Indoor Scenes" IROS 2023.

arxiv

The VL-Grasp is a interactive grasp policy combined with visual grounding and 6-dof grasp pose detection tasks. The robot can adapt to various observation views and more diverse indoor scenes to grasp the target according to a human's language command by applying the VL-Grasp. Meanwhile, we build a new visual grounding dataset specially designed for the robot interaction grasp task, called RoboRefIt.

Download the RoboRefIt Dataset

You can download RoboRefIt from Google Drive. The data directories should like this:

RoboRefIt/
├── data/
│   └── final_dataset/
│       ├── train
│       ├── testA
│       └── testB

The homepage of the RoboRefIt dataset is at RoboRefIt. More details and statistics information about the dataset can be found in the homepage.

Requirements

References：

python=3.7.16
torch=1.7.1+cu110
torchvision=0.8.2+cu110
torchaudio=0.7.2

And others:

cd RoboRefIt
pip install -r requirements.txt

cd GraspNet
pip install -r requirements.txt

cd GraspNet/knn
python setup.py install 

cd GraspNet/pointnet2
python setup.py install

Training

There are two stages of model training.

Visual Grounding Network

Training with the RoboRefIt dataset. Please download the dataset and allocate the dataset parameters at "./RoboRefIt/main_vg.py".

cd RoboRefIt
sh train_roborefit.sh

Thansks for the RefTR model.

6-Dof Grasp Pose Detection Network

Training with the GraspNet-1Billion dataset. Please download the dataset and allocate the dataset parameters at "./GraspNet/train.py".

cd GraspNet
sh train.sh

Thansks for the FGC-GraspNet model.

Model

There are Google Drive links, checkpoint_best_r50.pth [https://drive.google.com/file/d/1HJQKnuiG5J02PZBQJc2KsW3RozmJYH-L/view?usp=sharing] and checkpoint_fgc.tar [https://drive.google.com/file/d/1x4e23njqi4A_S_LlZPCUHjHT9CqFPzkc/view?usp=sharing].

Demo

python main.py

BibTeX

@inproceedings{lu2023vl,
  title={VL-Grasp: a 6-Dof Interactive Grasp Policy for Language-Oriented Objects in Cluttered Indoor Scenes},
  author={Lu, Yuhao and Fan, Yixuan and Deng, Beixing and Liu, Fangfu and Li, Yali and Wang, Shengjin},
  booktitle={2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={976--983},
  year={2023},
  organization={IEEE}
}