This is the PyTorch implementation of the paper Learning to Grasp (L2G).
Learning to Grasp (L2G) is an efficient end-to-end learning strategy to generate 6-DOF parallel-jaw grasps starting from a partial point cloud of an object. Our approach does not exploit any geometric assumption, it is instead guided by a principled multi-task optimization objective that generates a diverse set of grasps by combining contact point sampling, grasp regression, and grasp evaluation.
(click here to watch higher-res, full-length version)
This code is tested with Python 3.7.12, PyTorch 1.8.0 and CUDA 11.1 (system-wide).
Make sure to have the following environment variables (CUDA_HOME, CUDA_ROOT, LD_LIBRARY_PATH, PATH) properly set before installing pointnet2 modules. GCC and CMAKE are also required.
SYSTEM:
CUDA 11.1 (system-wide)
gcc-7.3.1
cmake-3.18.4
CONDA ENV:
torch: 1.8.0 # torch.__version__
CUDA: 11.1 # torch.version.cuda
CuDNN: 8005 # torch.backends.cudnn.version()
conda env create --name L2G_env --file=requirements.yml
pip install --upgrade https://github.com/unlimblue/KNN_CUDA/releases/download/0.2/KNN_CUDA-0.2-py3-none-any.whl
cd pointnet2;
python setup.py install;
cd ..
git submodule init
git submodule update
cd GPNet-simulator; pip install -e .; cd ..
cd grasp-evaluator; python setup.py install; cd ..
Use the script: download_data.sh to download the datasets (requires ~11GB disk space once unzipped). For a more detailed description of the datasets, please refer to their individual README files.
sh download_data.sh;
.
├── README.md
└── data
├── ShapeNetSem-8
│ └── ...
├── YCB-76
│ └── ...
├── YCB-8
│ └── ...
└── download_script.sh
Use the provided bash scripts to train and evaluate L2G models. Training is performed on ShapeNetSem-8, evaluation (rule-based + simulation) is performed on ShapeNetSem-8, YCB-8, YCB-76 test sets.
conda activate L2G_env
source run_l2g_pn2.sh
conda activate L2G_env
source run_l2g_deco.sh
Our code release is based on the following works
This work is partially funded by CHIST-ERA under EPSRC grant no. EP/S032487/1.
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。