VideoChat3 and RoboTTT Use Separate Transformer Backbones for Video Chat and Robot Policies

The prevailing trend in multimodal foundation models is to collapse vision, language, and action into...