nkalavak / dVRL

Contains Reinforcement Learning environments for the da Vinci Surgical System

RL Environments for the da Vinci Surgical System

Arxiv paper: https://arxiv.org/abs/1903.02090

Video link : https://www.youtube.com/watch?v=xu4sqrO_2AY

System set up, note that only linux is supported:

1. Requires Docker: https://docs.docker.com/install/linux/docker-ce/ubuntu/
2. Requires NVIDIA Container Runtime for Docker: https://github.com/NVIDIA/nvidia-docker
3. Enable GUI for Docker containers: http://wiki.ros.org/docker/Tutorials/GUI
4. Run bash script to build Docker Images: bash build_dockers.sh
5. Python packages: pip install transforms3d docker gym matplotlib

To speed up the simulation during training, V-REP can be launched within the docker in hidden mode.

To turn this on modify the last line in dVRL_simulator/environments/<reach/pick>_ee_dockerfile/Dockerfile. Add the "-h" flag in the final line:

CMD /app/V-REP/vrep.sh -h -s -q /app/scene.ttt

Test to check if NVIDIA Runtime can be used:

docker run --gpus all --rm nvidia/cuda:9.0-base nvidia-smi

Known issues and fixes:

1. User --user 1000 instead of username if name not found in password file

2. $ docker run --env="DISPLAY" --volume="/tmp/.X11-unix:/tmp/.X11-unix:rw" --gpus all vrep_ee_reach

QStandardPaths: XDG_RUNTIME_DIR not set, defaulting to '/tmp/runtime-root' No protocol specified QXcbConnection: Could not connect to display :0 /app/V-REP/vrep.sh: line 33: 13 Aborted (core dumped) "$dirname/$appname" "${PARAMETERS[@]}"

Fix: If you are not running docker as sudo, use the following: Run $xhost local:root

or try $xhost +

Current issue:

Specifying --runtime nvidia seems to be causing issues

docker run --env="DISPLAY" --volume="/tmp/.X11-unix:/tmp/.X11-unix:rw" --runtime=nvidia vrep_ee_reach