Docker

Docker provides a convenient way to package software into standardized units for development, shipment and deployment. See the official website for more details about Docker. NVIDIA Container Tookit enables users to build and run GPU accelerated Docker containers.

First, install nvidia-docker v2 following the official instructions. It is recommended to complete post-install steps for Linux.

To verify the installation:

# You should be able to run this without sudo.
docker run hello-world

Run ManiSkill2 in Docker

We provide a docker image (haosulab/mani-skill2) and its corresponding Dockerfile.

docker pull haosulab/mani-skill2
docker run --rm -it --gpus all haosulab/mani-skill2 python -m mani_skill2.examples.demo_random_action

Run GUI Applications

To run GUI applications from the docker container (the host is attached with a display), you need to add extra options to the docker run command:

# Allow local X11 connections
xhost +local:root
# Run ManiSkill2 docker image with the NVIDIA GPU
docker run --rm -it --gpus all \
    -v /tmp/.X11-unix:/tmp/.X11-unix -e DISPLAY \
    haosulab/mani-skill2 \
    python -m mani_skill2.examples.demo_manual_control -e PickCube-v0 --enable-sapien-viewer

To run GUI applications on a headless server, we present a solution based on x11vnc and fluxbox.

# https://www.richud.com/wiki/Ubuntu_Fluxbox_GUI_with_x11vnc_and_Xvfb
docker run --rm --gpus all -p 5900:5900 \
    haosulab/mani-skill2 \
    apt update && bash -c "apt install -yqq x11vnc fluxbox && x11vnc -create -env FD_PROG=/usr/bin/fluxbox  -env X11VNC_FINDDISPLAY_ALWAYS_FAILS=1 -env X11VNC_CREATE_GEOM=${1:-1920x1080x16} -gone 'pkill Xvfb' -nopw"

Then, forward the port of VNC (5900 by default) to the local host. On your local machine, install a VNC viewer and connect to the localhost port(e.g. localhost:5900).

References: