At a CIIE embodied-intelligence industry forum, Moushen Intelligence released its next-generation embodied-native model 'Moushen Shenwu,' an embodied-intelligence infrastructure 'Moushen ShenShu,' and a long-horizon memory home world model FWM.
Shenwu generates actions in one step with a vision-action hierarchical overlap, accelerating inference up to 24.84x, and requires only about 13.4 million additional trainable parameters for lightweight calibration. The company also set up a joint lab with Fudan University and formed an industry consortium with 42 organizations to advance general embodied brains.
