Reconstruct the moves and rotations of a human Rubik's cube solve from ordinary webcam video — no Bluetooth smart cube, no special hardware.
The core idea: the cube's structure is known in advance, so the learned models only ever locate points or read colors, while all 3D reasoning and reconstruction of unobserved moves is done by deterministic geometry and combinatorics.
Status: early scaffold (M0). The project is built one milestone at a time.
Requires Python 3.11+.
git clone https://github.com/IshPen/cube-tracker.git
cd cube-tracker
python -m venv .venv
# Windows: .venv\Scripts\activate
# macOS / Linux: source .venv/bin/activate
pip install -e ".[dev]"
pre-commit install
ruff check . && mypy && pytest