HumanEgo: Egocentric Manipulation Videos for Zero-Shot Robot Learning

HumanEgo is a dataset of human egocentric manipulation videos recorded with Project Aria (Gen 1) glasses. The dataset is paired with Meta MPS annotations, including SLAM trajectories and hand tracking, and provides full preprocessing output such as per-frame RGB and object/arm data. It was created by Leo-TX and last updated on June 8, 2026.

Use Cases

• Training robot manipulation policies based on human egocentric video demonstrations.
• Developing zero-shot robot learning algorithms using annotated SLAM trajectories and hand tracking.
• Benchmarking computer vision models for object and arm pose estimation from egocentric video.
• Researching human-robot interaction by analyzing natural human manipulation sequences.

Strengths

• Includes paired Meta MPS annotations (SLAM trajectories and hand tracking).
• Provides full preprocessing output, including per-frame RGB and object/arm data.
• Recorded with Project Aria (Gen 1) glasses, suggesting high-quality sensor data.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

Leo-TX on Hugging Face.

Collection Method

Recorded with Project Aria (Gen 1) glasses and annotated with Meta MPS.

Freshness

Last updated 2026-06-08 22:20:00; freshness should be verified.