计算机视觉

Think Like a World Model, Act Like a VLA: Distilling World-Model Representations into Compact Robot Policies

Vision-Language-Action (VLA) models map observations to actions with no objective that accounts for how the world respon