World-action models and the push for robots that adapt to new tasks

Latent Space describes Dyna Robotics’ Jason Ma explaining how world-action models predict the future to guide physical robot actions. He says training on over a million hours of human video lets robots adapt to new tasks with as little as 45 minutes of demonstrations. The discussion also covers synthetic data, self-improving models, hour-long autonomous workflows and robot deployments at Din Tai Fung.