Robot End-Effector and Hand State Data for Cloth Manipulation

Unitreerobotics provides a dataset of robot observations for cloth manipulation tasks. It contains end-effector states computed via forward kinematics and finger states for both hands, with specific details for the Inspire Hand. The dataset was last updated in March 2026.

Use Cases

• Train models to predict `observation.state.ee_state` from other sensor inputs for state estimation.
• Develop reinforcement learning policies for bimanual cloth manipulation using the 12-dimensional `observation.state.ee_state` and `observation.state.hand_state` as observations.
• Analyze the correlation between finger aperture (`observation.state.hand_state` ranging 0.0 to 1.0) and end-effector pose during cloth grasping tasks.
• Benchmark forward kinematics algorithms against the provided `observation.state.ee_state` data which includes waist contribution.

Strengths

• End-effector state data includes contributions from the robot's waist, providing a complete kinematic chain.
• Hand state data is provided for both hands with specific, documented ranges (e.g., 0.0–1.0 for Inspire Hand).

Limitations

• Unknown row count and dataset size prevent assessment of statistical power.
• Dimensionality of `observation.state.hand_state` varies by hand type, requiring careful handling.
• Sample data and column list are unavailable, complicating initial exploration.

Provenance

Source

Unitreerobotics

Collection Method

Collected from robot observations during cloth manipulation tasks; end-effector states computed via forward kinematics.

Freshness

Last updated March 27, 2026.