shubhxho2026mit
SPEALL — Synthetic Egocentric Manipulation Dataset
A multimodal egocentric manipulation dataset generated using Odyssey, a neural world model that streams physically consistent video in real time. SPEALL provides synthetic training data for egocentric manipulation policies and world models as a scalable alternative to costly real-world demonstrations.
Downloads30
Episodes100
Why This Matters for Physical AI
SPEALL demonstrates how neural world models can serve as scalable synthetic data sources for training embodied AI policies, addressing the cost and scalability limitations of real-world robot demonstrations while maintaining reasonable contact realism through learning from real-world video.
Technical Profile
- Modalities
- rgblanguage
- Environment
- simulation
- Task Types
- manipulation
- Episodes
- 100
- Data Format
- WebDataset
- Annotation Types
- language_instructions
- License
- mit
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack