InternW0: A Foundational Physical World Model for Efficient Real-World Interactions
Shanghai AI Laboratory introduces InternW0, a physical world model that jointly predicts visuals and continuous robot control.
InternW0 is the first model in Shanghai AI Laboratory's InternW physical world model series, combining omnimodal interfaces, asynchronous multi-frequency processing, and local physical modeling under partial observations. A high-capacity video expert and a faster lightweight action expert jointly learn visual dynamics and continuous robot control with flow matching, reusing layerwise key-value context for new observations. Training uses about 7,200 hours of heterogeneous robot and egocentric data, including the 275-hour EgoLab laboratory set. Evaluations cover simulation plus a 15-stage metal-organic framework synthesis and 5-stage contact- and force-aware pipetting.
- Asymmetric video-action architecture uses flow matching
- Reuses layerwise key-value context instead of regenerating futures
- Trained on about 7,200 hours, including 275-hour EgoLab
- Evaluated on MOF synthesis and force-aware pipetting
Full article196 words · extracted from huggingface.co · click to collapse
Physical intelligence requires more than predicting how the world may evolve: predictions must remain actionable as the world continues to change. We introduce InternW0, the first instantiation of the InternW physical world model series from Shanghai AI Laboratory, built around omnimodal interfaces, asynchronous multi-frequency processing, and local physical modeling under partial observations and external influences. InternW0 jointly learns future visual dynamics and continuous robot control through an asymmetric video--action architecture with flow matching. A high-capacity video expert provides longer-horizon predictive context, while a lightweight action expert operates at a faster timescale. Instead of regenerating the future for every action update, InternW0 reuses layerwise K/V and adapts it to newly observed states through observation-conditioned context routing. Domain-specific interfaces and soft prompts support heterogeneous embodiments, while contact-aware post-training incorporates force and tactile signals for contact-rich manipulation. We train InternW0 on approximately 7,200 hours of heterogeneous robot and egocentric data, including EgoLab, a 275-hour real-laboratory egocentric dataset. Evaluation spans simulation benchmarks and real-world scientific tasks, including a 15-stage metal--organic framework synthesis workflow and 5-stage contact- and force-aware dexterous manipulation for general-purpose quantitative pipetting. These results advance scalable, asynchronous, and science-native physical world models for universal and efficient real-world interactions.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.27656