Docker

Prerequisites

Docker with NVIDIA Container Toolkit (nvidia-docker2)
Access to nvcr.io/nvidia/isaac-lab:2.2.0 (base image)

Build

# Uses git short SHA as tag by default
./docker/build_docker.sh

# Custom tag
./docker/build_docker.sh my-tag

# Build and push to registry
./docker/build_docker.sh --push

# Custom tag and push
./docker/build_docker.sh my-tag --push

The build context is the repo root. A .dockerignore at the repo root excludes .git/, build artifacts, and non-essential directories to keep the context small. Code and assets/ are baked into the image. Each assets/ subdirectory is a separate layer for better caching — code changes don't invalidate asset layers.

Run

# Interactive shell with display/GUI forwarding (X11, cache + repo mounts)
./docker/run_docker.sh

# Or specify a custom tag
./docker/run_docker.sh my-tag

Running a single command

docker run --rm \
    --gpus all \
    --network=host \
    --entrypoint /workspace/isaaclab/_isaac_sim/python.sh \
    -e ACCEPT_EULA=Y \
    robolab:<tag> \
    <script.py> [args...]

Running with display (for GUI/rendering)

docker run --rm -it \
    --gpus all \
    --network=host \
    --entrypoint /bin/bash \
    -e ACCEPT_EULA=Y \
    -e DISPLAY=$DISPLAY \
    -v /tmp/.X11-unix:/tmp/.X11-unix \
    robolab:<tag>

What's in the image

Base: nvcr.io/nvidia/isaac-lab:2.2.0 (Isaac Lab + Isaac Sim + Python)
Code: robolab/, scripts/, examples/, tests/
Assets: assets/ (~6.5GB)
Python packages: Everything in requirements.txt, installed via pip install -e .
System tools: htop, nvtop, tmux, vim, git-lfs, zip