Cable Box: Robot Teleoperation Dataset with 61 Episodes

61 teleoperated episodes recorded for robot learning, totaling 100,143 frames at 30 frames per second. The dataset includes 26-dimensional joint position states and images from three camera views: head, left wrist, and right wrist. It was created by author 120ft and published on the Hugging Face platform.

Use Cases

• Train imitation learning models based on the teleoperated demonstration episodes.
• Develop visuomotor control policies based on the synchronized multiview camera observations.
• Benchmark robot learning algorithms on a dataset of real-world manipulation trajectories.
• Analyze robot state-action sequences based on the 26-dimensional joint position data.

Strengths

• Contains 61 distinct demonstration episodes, providing multiple task executions.
• Includes 100,143 frames of synchronized sensor data, offering substantial temporal coverage.
• Provides multimodal observations combining state data (26-dim joint positions) and images from three camera angles.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Recorded with the luxo-summer-camp teleop stack.

Freshness

Last updated 2026-07-02 23:33:11; freshness should be verified.