Humanoid Robot End-Effector States for Laundry Task

Unitree Robotics recorded state data from a humanoid robot performing a clothes-loading task. The dataset captures end-effector poses and hand finger states during manipulation. It was published by unitreerobotics in March 2026.

Use Cases

• Train reinforcement learning agents for bimanual manipulation using the 12-dimensional observation.state.ee_state feature.
• Model the dynamics of hand closing actions using the observation.state.hand_state values ranging from 0.0 (open) to 1.0 (close).
• Benchmark forward kinematics (FK) models against the provided end-effector states which include waist contribution.
• Analyze coordination between left and right end-effectors via the concatenated pose data in observation.state.ee_state.

Strengths

• Data includes forward kinematics-computed end-effector states for both arms.
• Hand state values are normalized to a consistent 0.0 to 1.0 range for the Inspire Hand.

Limitations

• Sample size and total number of rows are unknown.
• The dataset appears focused on a single, specific task (putting clothes in a washer).
• Dimensionality of observation.state.hand_state varies (12 or 2) depending on hand type, which may complicate processing.

Provenance

Source

unitreerobotics

Collection Method

Recorded from robot sensors during task execution, with end-effector states computed via forward kinematics.

Freshness

Last updated March 2026.

Geography

Region 'us' indicated by platform tags.