Contact-GraspNet Repository

NOTE: This section is a not part of the official document. If your looking for orignal document, go down to find the official contents or go to the orignal repository.

1. Prerequisite

1.1 Using conda env

Before using conda download the dev_ros branch only in your ros ws/src and catkin build.

cd your_ros_workspace/src

git clone -b dev_ros https://github.com/deyakovleva/contact_graspnet.git

cd ../

catkin build

source devel/setup.bash

Create the conda env

conda env create -f cgp_env_last.yml

Troubleshooting

• Recompile pointnet2 tf_ops:

sh compile_pointnet_tfops.sh

Activate conda env

conda activate contact_graspnet_env

2. Download Models and Data

Model

Download trained models from https://drive.google.com/drive/folders/1tBHKf60K8DLM5arm-Chyf7jxkzOr5zGl and copy them into the checkpoints/ folder.

Test data

Download the test data from https://drive.google.com/drive/folders/1v0_QMTUIEOcu09Int5V6N2Nuq7UCtuAA and copy them them into the test_data/ folder.

3. ROS Server Interface

3.1 grasp_planner (ContactGraspNetPlanner)

3.1.1 Service Request Messages

• color_image (sensor_msgs/Image)
  • Color image. It is not used for generating grasps.
• depth_image (sensor_msgs/Image)
  • Detph image
• camera_info (sensor_msgs/Camerainfo)
  • Depth camera intrinsic for deprojecting depth image to point clouds.
• segmask (sensor_msgs/Image)
  • Object instance segmentation for grasp filltering.

3.1.2 Service Reponse Messages

• grasps (ContactGrasp[])
  • List of generated grasps.

3.1.3 Arguments

• ckpt_dir (str)
  • Contact-GraspNet checkpoint directory.
  • Default: checkpoints/scene_test_2048_bs3_hor_sigma_001
• z_min (double)
  • Z min value threshold to crop the input point cloud.
  • Default: 0.2
• z_max (double)
  • Z max value threshold to crop the input point cloud.
  • Default: 1.8
• local_regions (bool)
  • Crop 3D local regions around given segments.
  • Default: False
• filter_grasps (bool)
  • Filter grasp contacts according to segmap.
  • Default: False
• skip_border_objects (bool)
  • When extracting local_regions, ignore segments at depth map boundary.
  • Default: False
• forward_passes (int)
  • Run multiple parallel forward passes to mesh_utils more potential contact points.
  • Default: 1
• segmap_id (int)
  • Only return grasps of the given object id
  • Default: 0

4. Launch ROS node

4.1 How to use my_node.py

Start Grasp Planner Server Node

roslaunch contact_graspnet_planner grasp_planner.launch

Change path to global_config in 253 line in my_node.py. Start my_node.py

cd contact_graspnet/
python3 my_node.py

In line 210 the argument can be changed (depth_ros - for full depth image and depth_crp_ros - for cropped depth image)

Start rviz with aligned_depth_to_color

roslaunch realsense2_camera rs_aligned_depth.launch

Call service for grasps generation

rosservice call /responce "{}"

Citation

@article{sundermeyer2021contact,
  title={Contact-GraspNet: Efficient 6-DoF Grasp Generation in Cluttered Scenes},
  author={Sundermeyer, Martin and Mousavian, Arsalan and Triebel, Rudolph and Fox, Dieter},
  booktitle={2021 IEEE International Conference on Robotics and Automation (ICRA)},
  year={2021}
}