Humanoid-X: Human Pose and Action Data for Robotic Control

USC-PSI-Lab's Humanoid-X dataset supports the paper "Learning from Massive Human Videos for Universal Humanoid Pose Control". It includes released text descriptions, humanoid keypoints, and humanoid actions data, with only part of the human poses data fully released. The dataset was last updated on January 14, 2025.

Use Cases

• Training humanoid pose control models based on the released keypoints data.
• Generating text descriptions for human actions based on the provided text data.
• Developing multimodal robotics systems that integrate text, keypoints, and action data.

Strengths

• Data is officially released for a published research paper.
• Includes multiple data modalities: text descriptions, keypoints, and actions.
• Last updated on 2025-01-14, indicating recent maintenance.

Limitations

• Only part of the human poses data is released, as stated in the description.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset size are unknown, which may limit suitability assessment.

Provenance

Source

USC-PSI-Lab

Collection Method

Created for the paper "Learning from Massive Human Videos for Universal Humanoid Pose Control".

Time Range

null

Freshness

Last updated 2025-01-14 19:29:12.

Geography

null