H2RB bridges human-to-robot VLA transfer
Original title: Human-to-Robot Transfer via Bridging Representation and Action Space in VLAs
H2RB co-trains VLAs on unpaired human and robot data for 51.8 object transfer success.
Pretrained VLAs struggle adapting to new robots. H2RB uses geometry-anchored alignment and decoupled-decoder tokenizer to bridge embodiment gaps. It improves transfer from human demos across real-world tasks and object shifts.
- Object transfer improves to 51.8
- Skill transfer reaches 62.7
- Out-of-distribution transfer to 45.3
- Reduces need for target-robot data
- Organizations
- Hugging Face
- AI models
- H2RB
From the original article
1 min read at huggingface.co
Pretrained Vision-Language-Action (VLA) policies provide a strong starting point for deployment, but adapting them to a target robot and environment still requires costly embodiment-specific demonstrations.
By Junhyeong Park, Wook Jung, Taeyoung Kim · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.