CS492_Intelligence_Contest

forked from Deep Grasping ROS (Seunghyeok Back)

• Project member: Chanyoung Ahn, Taeyeoung lee
• Project Result : [Video], [Image]

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

Contact-GraspNet Framework

Deep_grasping/src

Contact_Graspnet_server.py (for ROS)
/Contact_graspnet
 - Contact_graspnet_client.py (for ROS)
/uoais
 - launch/uoais_rs_d435.launch (need to modify name space)
 -(TOBO) uoais_client.py
(TODO)namespace 

Env setup

1. Install Repository (CS492_Intelligence_contest/src)

1. git clone 'CS_492_intelligence_contest' or 'gist-ailab/deep-grasping'
2. 'cd src', git clone 'contact_graspnet','uoais'

git clone https://github.com/cold-young/contact_graspnet.git
git clone https://github.com/cold-young/uoais.git

• Enviroment setup(1~7) + Test "ROS nodes 1.Realsense D-435" ..(나중에 namespace 변경 필요할듯)

• (https://github.com/gist-ailab/uoais#environment-setup)

3. Install easy_tcp_python2_3, open3d-ros-helper (Use conda env!!!)

• https://github.com/SeungBack/open3d-ros-helper (env py2.7, pip2 install~)
• https://github.com/SeungBack/easy_tcp_python2_3 (env py2.7 or 3.6~, pip install~)

2. Create conda enviroment

1. contact_graspnet_env(for contact_graspnet_client.py)

• python 3.7, tensorflow 2.2, CUDA 10.1, CUCNN 7.6.0
• yml file location: https://github.com/SeungBack/contact_graspnet
• Install : easy_tcp_python2_3

conda env create -f contact_graspnet_env.yml
pip install pickle-compat
pip install opencv-python
pip install easy-tcp-python2-3

2. uoais(for uoais.py : 작동 O)

• Enviroment: python 3.7, pytorch 1.8.0+cu111, torchvision 0.9.0+cu111, CUDA 11.1, detectron2 v0.6+cu111+torch 1.8.0

conda create -n uoais python=3.7
conda activate uoais
pip install torch==1.8.0+cu111 torchvision==0.9.0+cu111 torchaudio==0.8.0 -f https://download.pytorch.org/whl/torch_stable.html
pip install shapely torchfile opencv-python pyfastnoisesimd rapidfuzz

(please CUDA version check: nvcc -V, torch.version.cuda ) (If, CUDA version confilct => uoais setup.py build develop 실행 안됨 )

• install detectron2... (https://detectron2.readthedocs.io/en/latest/tutorials/install.html#requirements)

RUN

Robot(HAETAE)

• If you run in CPU machine(attaced Haetae..), try http://192.168.10.10:11311 and wifi off

Haetae RUN(+ wrist_camera)

export ROS_MASTER_URI=http://192.168.0.100:11311
export ROS_HOST_URI=http://192.168.0.24 ??????
sudo route add -net 192.168.10.10 netmask 255.255.255.255 gw 192.168.0.100
(or alias 'robot')
(navigation_ws(janghyuk/12_ob..detec..)
source ./robot.sh real manip
roslaunch pick_and_place_demo demo.launch sim:=false perception_src:=none

if you use zsh, robot.sh -> remove ~~ source .bash ~~ and put ~/.zshrc 'export $HOSTNAME = ${YOUR machine name}'

(To modify...)

export ROS_MASTER_URI=http://192.168.0.100:11311
sudo route add -net 192.168.10.10 netmask 255.255.255.255 gw 192.168.0.100
rostopic list

TODO: localhost:11311 -> haetae 
launch file 
bash file ....

RViz(Haetae ROS 연결 확인 후)

• If you want to use 'wrist_camera' topics, have to run 'demo.launch'

export ROS_MASTER_URI=http://192.168.0.100:11311
source navigation_ws/devel/setup.bash
rosrun rviz rviz

Contact graspnet server

• python == 2.7, with dependencies of rirolab/navigation_ws

roscd deep_grasping && python src/contact_grasp_server.py

Contact graspnet client

1. contact graps client.py

• python == 3.7 please create conda env(OK)

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

• Check grasp pose, orientation

rosservice call /get_target_grasp_pose      

2. uoais client.py (예제는 작동 됨)

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

We must make 'uoais_client.py' or other files \

2.1 For uoais example 1 (Image example)

python tools/run_on_OSD.py --use-cgnet --dataset-path ./sample_data --config-file configs/R50_rgbdconcat_mlc_occatmask_hom_concat.yaml

2.2 For uoais example 2 (use ROS RViz)
We use ROS1 version + python3:: cv_bridge Problem
If you make catkin_ws 'catkin_make', You need to create New workspace

Create 'cvbridge_build_ws Link

sudo apt-get install python3-pip python-catkin-tools python3-dev python3-numpy
sudo pip3 install rospkg catkin_pkg

mkdir -p ~/cvbridge_build_ws/src

catkin config -DPYTHON_EXECUTABLE=/usr/bin/python3 -DPYTHON_INCLUDE_DIR=/usr/include/python3.6m -DPYTHON_LIBRARY=/usr/lib/x86_64-linux-gnu/libpython3.6m.so
# Instruct catkin to install built packages into install place. It is $CATKIN_WORKSPACE/install folder
catkin config --install

cd ~/cvbridge_build_ws/src
git clone -b melodic https://github.com/ros-perception/vision_opencv.git


apt-cache show ros-melodic-cv-bridge | grep Version
    Version: 1.13.0-0xenial-20180416-143935-0800
# Checkout right version in git repo. In our case it is 1.13.0
cd src/vision_opencv/
git checkout 1.13.0
cd ../../

# Build
catkin build cv_bridge

# Extend environment with new package
source install/setup.bash --extend

• Check your workspace

python
from cv_bridge.boost.cv_bridge_boost import getCvType 
exit()

Then, We can use this launch file

conda activate uoais
roslaunch uoais uoais_rs_d435.launch