Breaking the Vision-Action Shortcut: Latent Interface Training for Generalizable Robotics Foundation Models Paper • 2609.12641 • Published 4 days ago • 61