dmlab-gym 0.1.0

DeepMind Lab with Bazel 8, Python 3.13, and Gymnasium/Shimmy support

dmlab-gym

A fork of DeepMind Lab updated for Bazel 8, Python 3.13, and wrapped with Gymnasium via Shimmy.

What This Is

DeepMind Lab is a 3D learning environment based on id Software's Quake III Arena. This fork modernizes the build toolchain so it compiles and runs on current systems:

• Bazel 8 with --enable_workspace compatibility
• Python 3.13 C extension and package support
• Gymnasium/Shimmy wrapper for standard RL interfaces

The engine builds inside a container (Podman or Docker). The resulting wheel installs into a host-side uv environment.

Platform Support

Platform	Status
Linux (x86_64)	Supported
macOS	Not supported
Windows	Not supported

Prerequisites

• uv (Python environment management)
• Podman or Docker (container builds)
• Python 3.13+
• OSMesa (runtime dependency for headless rendering) — installed automatically by dmlab-gym build if missing

Manual OSMesa install

Distribution	Command
Fedora / RHEL	sudo dnf install mesa-libOSMesa or mesa-compat-libOSMesa
Bluefin / immutable	sudo dnf install --transient mesa-compat-libOSMesa (resets on reboot)
Ubuntu / Debian	sudo apt install libosmesa6
Arch	sudo pacman -S mesa

Building

Quick Start

dmlab-gym build                        # build and install deepmind-lab
dmlab-gym build -o ~/my_output         # custom output directory
dmlab-gym build --no-install           # build only, skip install

This auto-detects Podman or Docker, builds the native extension inside a container, and installs the resulting wheel into your current environment.

Verify

python -c "import deepmind_lab; print('OK')"

Manual Build

If you prefer to run the steps yourself:

Podman:

podman build -t dmlab-builder -f Dockerfile.build .

mkdir -p /tmp/dmlab_pkg
podman run --rm \
    -v ./:/build/lab:Z \
    -v /tmp/dmlab_pkg:/output:Z \
    dmlab-builder \
    bash -c "
        bazel build -c opt //python/pip_package:build_pip_package \
            --define headless=osmesa && \
        ./bazel-bin/python/pip_package/build_pip_package /output
    "

Docker:

docker build -t dmlab-builder -f Dockerfile.build .

mkdir -p /tmp/dmlab_pkg
docker run --rm \
    -v ./:/build/lab \
    -v /tmp/dmlab_pkg:/output \
    dmlab-builder \
    bash -c "
        bazel build -c opt //python/pip_package:build_pip_package \
            --define headless=osmesa && \
        ./bazel-bin/python/pip_package/build_pip_package /output
    "

Project Structure

Directory	Description
engine/	ioquake3 game engine (GPLv2)
q3map2/	Map compilation tools (GPLv2)
assets/	DeepMind Lab assets (CC BY 4.0)
assets_oa/	Open Arena assets (GPLv2)
game_scripts/	Lua game/level scripts
python/	Python C extension and pip package
deepmind/	Core DeepMind Lab code
public/	Public API headers
third_party/	Vendored dependencies
dmlab_gym/	Python package (CLI, utilities)
bazel/	Bazel build configuration
testing/	Test utilities
examples/	Example agent code

License

This project is a derivative work of DeepMind Lab and is distributed under the same licenses:

• GPLv2 -- engine/, q3map2/, assets_oa/
• CC BY 4.0 -- assets/
• Custom academic license -- engine/code/tools/lcc/
• BSD 3-Clause -- q3map2/libs/{ddslib,picomodel}

See LICENSE for full details.

Attribution

Based on DeepMind Lab by DeepMind.

If you use DeepMind Lab in your research, please cite the DeepMind Lab paper.

Upstream Sources

DeepMind Lab is built from the following open source projects:

• ioquake3 -- game engine
• bspc -- bot route compilation
• GtkRadiant / q3map2 -- map compilation