Dyna Robotics has developed Dyna-2, a world-action model that has been pre-trained on a vast dataset of over one million hours of human video footage. This extensive training enables the model to learn general patterns and behaviors from human activity. The model's performance has been demonstrated through three key results, including the ability to scale effectively with increased data and transfer its knowledge to unseen robotic scenarios. Furthermore, the model has shown improved cross-embodiment generalization when trained with video co-training. This advancement in AI research could have implications for the development of more sophisticated robotic systems.