Egocentric Human-Humanoid Data for Manipulation Policy Co-Training

PH2D contains egocentric human-humanoid interaction data for co-training manipulation policies for humanoid robots. The dataset, created by RogerQi, was released in 2025 in conjunction with the associated research paper. Data is organized into folders by task and primarily stored in HDF5 files.

Use Cases

• Co-training robot manipulation policies using egocentric video and proprioceptive data from human-humanoid interactions.
• Benchmarking imitation learning algorithms on task-specific data organized in HDF5 files.
• Developing visuomotor control models from the egocentric perspective data for humanoid tasks.

Strengths

• Released in 2025, ensuring relevance to current robotics research.
• Data is structured by specific manipulation tasks for focused training.

Limitations

• Specific row counts, column names, and dataset size are unknown.
• Data format and organization details require visiting the external source page.

Provenance

Source

RogerQi on Hugging Face Datasets.

Collection Method

Collected egocentric human-humanoid interaction data, methodology detailed in the associated arXiv paper.

Freshness

Last updated on 2025-04-13.