Installation¶

A focused installation checklist. For the full learning path, start with the tutorial.

Requirements¶

Requirement	Version
Python	3.12 or newer
Dependency manager	`uv`
Simulator backend	`genesis-world>=0.4.7`
PyTorch	`torch>=2.8.0` through one `torch-*` extra

Syncing the environment¶

From the repository root:

uv sync --extra torch-cpu

Choose exactly one PyTorch extra:

Extra	Target
`torch-cpu`	CPU-only or non-NVIDIA development machines.
`torch-cu126`	NVIDIA driver compatible with CUDA 12.6 wheels.
`torch-cu128`	NVIDIA driver compatible with CUDA 12.8 wheels.
`torch-cu130`	NVIDIA driver compatible with CUDA 13.0 wheels.

Refresh an existing torch install:

uv sync --reinstall-package torch --extra torch-cu128

Running commands¶

The rest of these docs invoke the CLI as a bare genelab (and a bare python) — not uv run genelab.

Why not uv run?

uv run performs an implicit uv sync before every command. The PyTorch builds ship as mutually-exclusive extras (torch-cpu / torch-cu126 / torch-cu128 / torch-cu130) that are not part of the default sync set, so each uv run uninstalls and reinstalls torch and rewrites the extra selected above — slow, and it can silently flip the CUDA build. Use one of the two ways below instead.

Activate the venv (recommended)Use uv without re-syncing

Activate once per shell, then run bare commands as shown throughout the docs:

source .venv/bin/activate      # Windows: .venv\Scripts\activate
genelab --version

Prefer not to activate? Skip the implicit sync on every call with --no-sync:

uv run --no-sync genelab --version

(Anywhere the docs write genelab … or python …, read it as uv run --no-sync genelab ….)

Why uv sync and uv pip install keep the uv prefix

Only uv run is the footgun — it re-resolves the project and reconciles the venv before executing. uv sync is the one deliberate sync that selects the torch build, and uv pip install -e … is uv's pip-compatible installer: it installs into the active venv without re-resolving the project or touching extras, so it never reinstalls torch. Both are safe to keep.

Creating local caches¶

genelab cache

The command creates writable project-local cache directories and points XDG_CACHE_HOME and MPLCONFIGDIR at .cache/.

Verifying imports¶

python -c "import genelab; print(genelab.__version__)"
python -c "from genelab.lab import TaskCfg; print(TaskCfg.__name__)"
python -c "import torch; print(torch.__version__, torch.version.cuda)"
python -c "import genesis; print(genesis.__version__)"

Installing example extensions¶

The core package is a framework; tasks are registered by extensions.

uv pip install -e examples/inverted_pendulum
uv pip install -e examples/genelab_examples
genelab list tasks

Install Unitree only for the larger humanoid examples:

uv pip install -e examples/unitree