EgoWorld: Egocentric Bimanual Manipulation with 3D Hand Poses

EgoWorld is a LeRobot v3.0 dataset of egocentric bimanual manipulation from human demonstrations. It features world-frame 3D hand poses, full MANO hand meshes, camera trajectories, and dense depth maps. The dataset was created by haoyang-li and was last updated on March 30, 2026.

Use Cases

• Training Vision-Language-Action models based on egocentric video and 3D hand pose data.
• Developing world models for robotic manipulation based on camera trajectories and depth maps.
• Training imitation learning policies for bimanual tasks based on human demonstration data.
• Research in 3D hand pose estimation and reconstruction using MANO hand meshes.

Strengths

• Includes world-frame 3D hand poses, which are a key feature for spatial reasoning.
• Contains full MANO hand meshes, providing detailed hand shape and articulation data.
• Provides camera trajectories and dense depth maps, offering multi-modal scene information.

Limitations

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

Provenance

Source

haoyang-li on Hugging Face

Collection Method

Human demonstrations for egocentric bimanual manipulation.

Freshness

Last updated 2026-03-30 22:19:20; freshness should be verified.