LeRobot documentation we used

Links to the official LeRobot docs behind each step. The library itself is at github.com/huggingface/lerobot, and the docs are at huggingface.co/docs/lerobot.

Assembly

• Assemble SO-101Step-by-step build, motor setup and calibration.
• Feetech motorsThe servos and their bus.
• InstallationIncluding the Feetech extra the SO-101 needs.

Getting started

• LeRobot documentation homeOverview of the library and what it supports.
• InstallationEnvironment setup and the optional extras.
• Imitation learning on real robotsThe full teleoperate, record, train and evaluate workflow.

Hardware

• SO-101Assembly, motor IDs, calibration.
• SO-100The earlier version of the arm.
• Feetech motorsMotor bus and servo details.
• CamerasDiscovering and configuring OpenCV cameras.
• Integrate your own hardwareAdding a new robot type, as our MuJoCo robot does.

Datasets

• LeRobotDataset v3The on-disk format our recordings use.
• Streaming video encodingEncoding while recording.
• Dataset toolsEditing and merging datasets.

Policies and training

• ACTAction Chunking Transformer.
• SmolVLAA compact vision-language-action model.
• π0Physical Intelligence's policy.
• InferenceRunning a trained policy on the robot.

Kinematics and simulation

• Processors for robots and teleoperatorsIncludes RobotKinematics, used in our MuJoCo work.
• HIL-SERL simulationLeRobot's own simulation example.
• LeIsaacControl and train robots in simulation.

Other references

• LeRobot on GitHubSource code and issues.
• SO-ARM100Hardware design, bill of materials and the MuJoCo model we vendored.
• Hugging Face LeRobot orgDatasets and pretrained policies.
• RerunThe visualization tool used with --display_data.
• MuJoCoThe physics engine behind our simulator.