InternW0: A Foundational Physical World Model for Efficient Real-World Interactions
InternW0 couples future video prediction with fast robot control so actions can update without regenerating the whole forecast.
Shanghai AI Laboratory presents it as the first model in its InternW physical world model series. The system uses an asymmetric video-action design, with a larger video expert for longer-horizon context and a lightweight action expert running faster. It was trained on about 7,200 hours of robot and egocentric data, including the 275-hour EgoLab dataset. Evaluation includes simulation and real-world scientific tasks, such as multi-stage metal-organic framework synthesis and dexterous quantitative pipetting. HF Daily Papers' note
Shanghai AI Laboratory presents it as the first model in its InternW physical world model series. The system uses an asymmetric video-action design, with a larger video expert for longer-horizon context and a lightweight action expert running faster. It was trained on about 7,200 hours of robot and egocentric data, including the 275-hour EgoLab dataset. Evaluation includes simulation and real-world scientific tasks, such as multi-stage metal-organic framework synthesis and dexterous quantitative pipetting. HF Daily Papers' note
score 6