H2R-1M: Human-to-Robot Egocentric Video Dataset for Robot Pre-training

H2R-1M is a family of large-scale robot-centric video datasets generated via a visual data augmentation pipeline. The dataset transforms human egocentric manipulation videos into robot-augmented versions by replacing human hands with simulated robot arms using pose estimation and physically aligned rendering. It was created by author yaoxu789 and last updated on Hugging Face on 2025-05-24.

Use Cases

• Pre-training robot manipulation policies based on the augmented egocentric video data.
• Developing sim-to-real transfer methods based on the physically aligned rendering pipeline.
• Benchmarking video-based imitation learning algorithms on robot-augmented human demonstrations.

Strengths

• Dataset is described as 'large-scale'.
• Data is generated via a specific pipeline using accurate pose estimation and physically aligned rendering.
• Last update timestamp is explicitly provided: 2025-05-24 10:41:27.

Limitations

• Dataset size, row count, and file formats are unknown.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Collection Method

Generated via the H2R visual data augmentation pipeline, which transforms human egocentric videos.

Freshness

Last updated 2025-05-24 10:41:27.