Python project for controlling a HiWonder xArm to manipulate and solve a Rubik's Cube. Includes motion control, camera integration, and cube-solving policy execution.
- Dual-process architecture: Concurrent vision analysis and robot execution
- Computer vision: Real-time cube detection, color classification, and state tracking with rubikvision
- Learned policies: SmolVLA or ACT
- Real-time visualization: Rerun integration for live monitoring and debugging
# Clone with submodules
git clone --recursive https://github.com/yourusername/xarm-cuber.git
cd xarm-cuber
# Setup Python environment
conda create -n cube python=3.10
conda activate cube
# Install dependencies
pip install -e thirdparty/lerobot/src
pip install -e thirdparty/rubikvision# Run autonomous solver
python src/main.py \
--robot.cameras="$CAMERAS_CONFIG" \
--robot.type=xarm_follower \
--robot.port=/dev/ttyUSB0 \
--policy.path="$POLICY_PATH" \
--dataset.repo_id="$REPO_ID" \
--dataset.single_task="task" \
--display_data=true \
--resume=true \
--policy.chunk_size=20 \
--policy.n_action_steps=20
┌──────────────────┐ ┌──────────────────┐
│ Robot & │◄──── action ──────► │ Vision & │
│ Camera │ │ Planning │
│ Process │──── image/busy ───► │ Process │
│ │ │ │
│ - Camera capture │ │ - Cube detection │
│ - Policy control │ │ - State analysis │
│ - Action chunking│ │ - Solve planning │
│ - Robot commands │ │ - Action mapping │
│ - Task execution │ │ - Kociemba solve │
│ - Rerun logging │ │ - Rerun logging │
└──────────────────┘ └──────────────────┘
Flow: Camera → Vision Analysis → Action Planning → Robot Execution → Repeat
Communication: PyTorch shared memory (image, busy flag, action commands)
# Record demonstrations
bash scripts/record_demo.sh
# Train ACT policy
bash scripts/train_act.sh- LeRobot for the robotics framework
- Kociemba algorithm for optimal cube solving sequences