Brain
HuRo is an academic robotization pipeline and dataset that converts heterogeneous egocentric human videos into robot-aligned observations and retargeted actions for VLA pretraining. Primary sources are the Sep 2026 paper (arXiv 2609.10706) and project page; digest coverage via @robotsdigest is secondary.
Research model · Maturity: Research · Open source
Machine-readable surfaces
- Markdown mirror: /brains/huro.md
- RSS feed: /brains/huro/feed.xml
- JSON-LD: embedded in this page’s head
- REST API: /v1/brains/f8b8f37e-07dc-40b7-bf4b-66fa4d673378
- Revision history: /brains/huro/history
- Data documentation: /data
- Query this programmatically: Deploy MCP
Architecture
Egocentric human-video robotization (visual conversion + motion retargeting) into ALLEX-aligned VLA episodes; sources include EgoDex/EgoVerse/Ego4D/Ego10K/EPIC-Kitchens
Key facts
- Dataset / pipeline scale (paper-claimed)
- Paper (arXiv 2609.10706, project page): HuRo robotization pipeline converts egocentric human videos into robot-aligned observations and retargeted actions. Main dataset ~630k robotized episodes / ~142M frames (~1,317 hours at 30 fps) from EgoDex, EgoVerse, Ego4D, Ego10K, and EPIC-Kitchens, instantiated primarily on the ALLEX bimanual dexterous humanoid
- VLA real-robot gains (paper-claimed)
- Paper-claimed across four real-world ALLEX manipulation tasks: scaling HuRo pretraining raises overall task completion from 51.5% to 80.3% and OOD completion under spatial/visual shifts from 34.9% to 72.2%. Academic result, not a commercial product claim
- 630K episodes 142M frames ALLEX (paper abs)
- arXiv abs 2609.10706 (Jeong, Joo et al.; RLWRLD / Yonsei) describes HuRo robotization converting heterogeneous egocentric human videos into robot-aligned observations and retargeted actions. Paper claims about 630K robotized episodes and 142M processed frames from five sources (Ego4D, EPIC-Kitchens, EgoDex, EgoVerse, Ego10K), mainly for ALLEX bimanual dexterous robot. Treat scale as paper-claimed (arXiv)
- VLA ID/OOD completion scaling (paper abs)
- Across four real-world ALLEX manipulation tasks, the paper claims pretraining on increasing HuRo subsets raises overall completion from 51.5% to 80.3% and OOD completion under spatial/visual shifts from 34.9% to 72.2%, with ablations that visual robotization helps OOD and end-to-end pretraining with retargeted actions beats visual-only transfer. Treat percentages as paper-claimed (arXiv)
- Human-video robotization
- HuRo presents a pipeline for converting human videos into robot-aligned observation and action data for vision-language-action pretraining (HuRo).
- ALLEX humanoid support
- The HuRo research materials describe support for the ALLEX bimanual humanoid robot in the generated training data pipeline (HuRo paper).
Common questions
What is HuRo?
Is HuRo open source?
What type of AI is HuRo?
What is HuRo's maturity stage?
Which robots run on HuRo?
Sources (4)
- https://arxiv.org/abs/2609.10706 · 2026-09-18
- https://3587jjh.github.io/HuRo/ · 2026-09-18
- https://x.com/robotsdigest/status/2102807448826601634 · 2026-09-23
- https://arxiv.org/pdf/2609.10706.pdf
Methodology: Verified · 4 sources (no primary) · last reviewed 2026-10-09
Verification posture
Verified
Low confidence
Review state
Stable
Last reviewed 2026-10-09
Maturity + lifecycle
Maturity stage: research
Sources by quality tier
- 2
- preprint
- Preprint
- 2
- unclassified
- Unclassified source
The framework is documented at /methodology. Corrections at /corrections. Reviewer: DEPLOY editorial team.
Methodology surface for HuRo.Canonical ID f8b8f37e-07dc-40b7-bf4b-66fa4d673378