English · 中文

EgoSuite-Open10K

EgoSuite-Open10K is a human-centric egocentric video dataset for Embodied AI and Physical AI. The v1.0 release contains 10,000 selected clips, organized into 6 SKUs across the EgoStandard and EgoPro families.

Overview Video

Open the full MP4 video →

Dataset at a Glance

Item	Value
Release size	10,000 clips
Dataset families	EgoStandard / EgoPro
Dataset SKUs	6
Scene groups	7
Perspectives	Standard egocentric and wrist-view
Core annotations	Hand Pose / Body Pose / V7 Semantic
Intended use	VLA, world models, motion understanding, robot learning

SKU Product Matrix

Family	SKU	Capability	Clips	Recommended Use
EgoStandard	EgoStand	🖐️	8,400	Pre-training and rapid prototyping
	EgoStand-motion	🖐️ 🧍	500	Full-body motion analysis
	EgoFull	🖐️ 🧍 🧠	100	Multimodal and semantic experiments
EgoPro	EgoPro	🖐️ ⌚	750	Wrist-view manipulation
	EgoPro-motion	🖐️ 🧍 ⌚	200	Assembly and collaborative motion
	EgoProMax	🖐️ 🧍 🧠 ⌚	50	Richest published modality combination

Capability legend: 🖐️ Hand Pose · 🧍 Body Pose · 🧠 V7 Semantic · ⌚ Wrist View

Scene Distribution

Scene groups: Home · Hospitality · Retail · Sports · Logistics · Office · Industry

Modalities

Modality	EgoStand	EgoStand-motion	EgoFull	EgoPro	EgoPro-motion	EgoProMax
Egocentric video	✓	✓	✓	✓	✓	✓
Wrist-view video	—	—	—	✓	✓	✓
Hand Pose	✓	✓	✓	✓	✓	✓
Body Pose	—	✓	✓	—	✓	✓
V7 Semantic	—	—	✓	—	—	✓

Quick Start

Download the repository

pip install -U huggingface_hub
hf download lightwheel-ai/EgoSuite-Open10K --repo-type dataset --local-dir ./EgoSuite-Open10K

Load a configured subset

from datasets import load_dataset

dataset = load_dataset(
    "lightwheel-ai/EgoSuite-Open10K",
    "ego_pro_max",
    split="train",
    streaming=True,
)

sample = next(iter(dataset))
print(sample)

Available configuration names:

ego_stand
ego_stand_motion
ego_full
ego_pro
ego_pro_motion
ego_pro_max

Recommended Uses

• Vision-Language-Action models
• World models and future-state prediction
• Human-motion understanding
• Hand-object interaction
• Robot-policy pre-training
• Egocentric video understanding

Limitations

Pose and semantic annotations may contain estimation or labeling errors. Human demonstrations should not be treated as directly executable robot actions. Users should validate the data and resulting models for their own embodiment, environment and safety requirements.

License

This repository is currently marked as Apache License 2.0. Confirm the final data license, participant consent, provenance and responsible-use terms before public release.

Citation

@dataset{lightwheel_egosuite_open10k_2026,
  title     = {EgoSuite-Open10K},
  author    = {{Lightwheel AI}},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/lightwheel-ai/EgoSuite-Open10K}
}

Contact

• Website: https://lightwheel.ai/
• Contact: https://lightwheel.ai/contact
• Dataset: https://huggingface.co/datasets/lightwheel-ai/EgoSuite-Open10K