DECO-50: 50 Hours of Bimanual Dexterous Robot Manipulation with Tactile Sensing

DECO-50 comprises over 5 million frames of teleoperated data for bimanual dexterous manipulation with tactile sensing. The dataset includes 50 hours of data collected on real dual-arm robots across 4 scenarios and 28 subtasks. It was created by BAAI-Humanoid and was last updated on Hugging Face in February 2026.

Use Cases

• Training multimodal AI models for robotic control based on visual and tactile sensor data.
• Benchmarking dexterous manipulation algorithms on real-world bimanual tasks.
• Developing and evaluating imitation learning policies from teleoperated demonstrations.
• Researching the integration of tactile sensing for fine-grained object manipulation.

Strengths

• Contains 50 hours of teleoperated data, providing substantial training material.
• Includes over 5 million frames, offering high temporal resolution for action sequences.
• Covers a diverse set of 28 subtasks across 4 distinct manipulation scenarios.
• Features tactile sensing data collected from real dual-arm robots, enhancing realism.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific model training needs.
• Description metadata is limited; actual data quality and structure require manual inspection after download.

Provenance

Source

BAAI-Humanoid

Collection Method

Teleoperated data collection on real dual-arm robots.

Freshness

Last updated 2026-02-18 07:29:35; freshness should be verified.