FastWAM Authors

FastWAM base

Checkpoint

World-action model combining a video-generation expert and compact action expert.

Specifications

Architecture components: Flow-matching action expert
Architecture components: Video-generation expert
Backbone: Wan video expert with flow-matching action expert
Frameworks: LeRobot / PyTorch
Input modalities: Language instruction
Input modalities: RGB camera images
Input modalities: Robot state / proprioception
Languages: English
Model type: Robot policy
Model type: World-action model
Output action space: Continuous robot action chunks; future video is a training signal
Parameter count: 6.0207B parameters
Pipeline task: Robotics