This is the official PyTorch Implementation of IndusGrasp: A Novel Robotic Grasp Detection Method Using Synthetic Data for Disordered Industrial Scenarios
- The video is available in YouTube
sudo apt-get install libglfw3-dev libglfw3
sudo apt-get install libassimp-dev
pip install --pre --upgrade PyOpenGL PyOpenGL_accelerate
pip install cython cyglfw3 pyassimp==3.3 imgaug progressbar
pip install -r requments.txt`
-
Using the
tools/pybullet_dataset/pybullet_dataset.py
to create render poses with your own 3D mesh(xx.urdf) -
Using the
tools/creat_grasp.py
to create grasp representation with your own 3D mesh(xx.ply) -
Using the
dataset_generate.py
to create your dataset
python train.py
python test.py
- Our Method in dense scenarios
- GR-CNN, in sparse scenarios
- GraspNet, which doesn't work for industrial parts
- Our method can generate to unseen objects with different shapes and textures (Note: the network is only trained by Synthetic Data generated by object 1. )