Real-World Robot Teleoperation Demonstrations with Video and State Data

Microsoft's VITRA dataset contains real-world robot teleoperation demonstrations collected using a 7-DoF robotic arm with a dexterous hand and a head-mounted RGB camera. Each episode provides synchronized numerical state and action data alongside video recordings, intended for finetuning vision-language-action models for robotic manipulation.

Use Cases

• Finetune a vision-language-action model using synchronized video recordings and numerical state data for robotic manipulation tasks.
• Train a policy model on the numerical state and action data sequences from teleoperation demonstrations.
• Analyze the correlation between head-mounted RGB camera video frames and the corresponding robotic arm state data.

Strengths

• Includes synchronized video recordings and numerical state/action data per episode.
• Collected using a 7-DoF robotic arm equipped with a dexterous hand and head-mounted camera.
• Designed for finetuning in the VITRA project for scalable robotic manipulation.

Limitations

• Specific row count, column names, and dataset size are unknown.
• Sample data and file formats are unavailable for inspection.
• Geographic origin and temporal coverage of the demonstrations are not specified.

Provenance

Source

microsoft

Collection Method

Real-world teleoperation demonstrations collected using a robotic system.

Freshness

Last updated on 2026-02-09.