Train Kinova Gen3 arm to insert the plug into the board in Gazebo with deep reinforcement learning and imitation learning
We strongly recommend using docker to deploy environments, and we provide our environment zip pack docker.tar.gz.
tar -zxvf dockerup.tar.gz
docker load < dockerup.tar
sudo apt-get install x11-xserver-utils
xhost +
docker run -d \
-v /etc/localtime:/etc/localtime:ro \
-v /tmp/.X11-unix:/tmp/.X11-unix \
-e DISPLAY=unix$DISPLAY \
-e GDK_SCALE \
-e GDK_DPI_SCALE \
--gpus=all \
-e NVIDIA_DRIVER_CAPABILITIES=compute,utility \
-e NVIDIA_VISIBLE_DEVICES=all \
--name pytorch \
dockerup
docker exec -ti pytorch /bin/bash
You can also deploy the environment step by step by referring to kinova
cd /catkin_workspace
Load configuration file and open Gazebo to visualize, if you dont want visualization then add argument gazebo_gui:=false (Can't speed up training anyway).
source devel/setup.bash
roslaunch kortex_gazebo spawn_kortex_robot.launch gripper:=robotiq_2f_85 start_rviz:=false
Open a new terminal and use the following commands to start training
source devel/setup.bash
roslaunch kortex_examples moveit_example.launch