Being-H0.8 Context

Audience

Robot-learning developers

Workflow

Tactile-aware contact-rich manipulation

Specifications

• Model Architecture: Latent tactile world-action model
• Human Video Corpus: More than 500,000 hours of egocentric human video
• Shared Action Representation: TopoHand
• Pseudo Tactile Supervision: TactoHand contact maps and proximity fields
• Physical Demonstration Platforms: Two bimanual robot configurations
• Physical Demonstration Tasks: Ten contact-rich manipulation tasks
• Paper Status: Coming soon as of 28 July 2026

Tags

• Peer Group: Robot manipulation foundation model, Robot foundation model