Deep Grasping ROS

ROS wrapper for DNN based robotic grasping algorithms
Support GQ-CNN [paper] [code]
Support 6-DoF-GraspNet [paper] [code]
Support Contact-GraspNet [paper] [code]

To generate a grasping point, simply use get_target_grasp_pose service

rosservice call /get_target_grasp_pose

Before run the deep grasping nodes, launch camera nodes

ROS_NAMESPACE=azure1 roslaunch azure_kinect_ros_driver driver.launch color_resolution:=1440P depth_mode:=WFOV_UNBINNED fps:=5 tf_prefix:=azure1_

TODO

Cuurently, this repo may not work properly. I will modify it soon ...
python 2 --> python 3
remove TCP connections

GQ-CNN

RUN

GQ-CNN Server

rosrun deep_grasping_ros gqcnn_server.py

GQ-CNN Client

conda activate gqcnn
roscd deep_grasping_ros/src/gqcnn && python gqcnn_client.py

FC-GQ-CNN Client

conda activate gqcnn &&  \
    roscd deep_grasping_ros/src/gqcnn && \
    python fc_gqcnn_client.py

6-DoF-GraspNet

Setup

conda create -n 6dofgraspnet python=2
python -m pip install -r requirements.txt
cd src && git clone https://github.com/SeungBack/6dof-graspnet
conda activate 6dofgraspnet && pip install -r requirements.txt

RUN

Azure kinect node

ros27 && ROS_NAMESPACE=azure1 roslaunch azure_kinect_ros_driver driver.launch color_resolution:=1536P depth_mode:=WFOV_UNBINNED fps:=5 tf_prefix:=azure1_

6-dof-graspnet server

conda activate 6dofgraspnet \ 
    && roscd deep_grasping_ros/src \
    && python 6dgn_ros_server.py

6-dof-graspnet client

conda activate 6dofgraspnet \
    && roscd deep_grasping_ros/src/6dof-graspnet \
    && python -m demo.6dgn_client --vae_checkpoint_folder checkpoints/npoints_1024_train_evaluator_0_allowed_categories__ngpus_1_/

Contact-GraspNet

RUN

Robot

cw && ./franka.sh master 
ros27 && roslaunch franka_interface interface.launch load_gripper:=true load_demo_planning_scene:=false

Azure kinect node

ROS_NAMESPACE=azure1 roslaunch azure_kinect_ros_driver driver.launch color_resolution:=720P depth_mode:=WFOV_UNBINNED fps:=5 tf_prefix:=azure1_

rosrun tf static_transform_publisher 0.078997 0.001502 0.066975 0.697583 -0.005156 0.716441 -0.007980 /panda_hand /azure1_camera_base 100

contact graspnet server

ros27 && roscd deep_grasping_ros && python src/contact_grasp_server.py

contact graspnet client

conda activate contact_graspnet_env \
    && roscd deep_grasping_ros/src/contact_graspnet \
    && CUDA_VISIBLE_DEVICES=0 python contact_graspnet/contact_grasp_client.py --local_regions --filter_grasps

conda activate uoais \
    && roscd deep_grasping_ros/src/uoais \
    && CUDA_VISIBLE_DEVICES=1 python demo/uoais_client.py

For Unstructured Contact-GraspNet ''' un
&& roslaunch gail_camera_manager gail_camera_driver_un.launch

un
&& cd ~/catkin_ws/src/deep-grasping/src && python contact_grasp_server.py

uncg
&& cd /home/ailab/catkin_ws/src/deep-grasping/src/contact_graspnet/contact_graspnet && python contact_grasp_client.py

un
&& cd ~/catkin_ws/src/deep-grasping/src/uoais && python demo/uoais_client.py

un
&& rosservice call /get_target_grasp_pose '''

Authors

Seunghyeok Back seungback

gist-ailab/deep-grasping