ALOHA 2 Bimanual Robot Teleoperation Demonstrations

ALOHA-Cosmos-Policy contains real-world human-teleoperated bimanual manipulation demonstrations recorded on the ALOHA 2 robot platform. NVIDIA created this dataset to train the Cosmos-Policy-ALOHA-Predict2-2B model, with data collected at a control frequency of 25 Hz. The dataset was last updated on the Hugging Face platform in January 2026.

Use Cases

• Train imitation learning policies using teleoperated video demonstrations of bimanual tasks.
• Benchmark robot policy performance on real-world manipulation sequences recorded at 25 Hz.
• Analyze human-teleoperated control strategies for the ALOHA 2 platform's two ViperX 300 S arms.
• Develop multimodal models that integrate visual observations from the demonstrations with robot action data.

Strengths

• Data originates from a real-world robotic platform (ALOHA 2) providing practical demonstrations.
• Recordings are captured at a specified control frequency of 25 Hz for temporal consistency.

Limitations

• Specific dataset size, row count, and column details are unknown, hindering scalability assessment.
• The geographic and temporal coverage of the data collection is not specified, limiting reproducibility analysis.

Provenance

Source

NVIDIA, hosted on Hugging Face.

Collection Method

Real-world human-teleoperated demonstrations collected on the ALOHA 2 bimanual robot platform.

Time Range

null

Freshness

Last updated on the platform in January 2026.

Geography

null