Installation and first reproduction#
Historical demo environment#
The released setup reproduces the original pinned environment, not a latest-stable comparison environment. #41 tracks the new qualification. Upgrading packages does not preserve an old validation claim automatically.
On Linux x86_64, install uv, then:
git clone https://github.com/huangkiki/Dexlab.git
cd Dexlab
bash scripts/setup.sh
bash demos/apple-stem-grasp/run.sh --backend mujoco
# Or use official SuperDex FP64
bash demos/apple-stem-grasp/run.sh --backend superdex
Add --headless without a display. SDF preparation and planning take time; a visible window does not establish physics acceptance. No model API key is required. Full experiments and bulk rendering run on the experiment server; desktop work is lightweight and resource-bounded.
Evidence and acceptance#
Retain configurations, actual joint/object states, native contacts, independent scores, logs and continuous close-ups. Acceptance requires complete lift, support, retention, penetration and momentum checks, not a final pose or GIF.
Detailed installation · Commands and output directories · Asset terms and provenance
Build documentation only#
This environment does not install DexLab, PyTorch, MuJoCo or robot assets:
uv venv --python 3.12 .venv-docs
uv pip install --python .venv-docs/bin/python -r docs/site/requirements.txt
.venv-docs/bin/python scripts/build_docs.py
.venv-docs/bin/python -m http.server 8000 --directory docs/_build/html
Open http://localhost:8000; Chinese is the default with an English switch. Warnings fail the build. No physics is executed.