VIDEAS: Distilling Explicit Action Semantics from Demonstration Videos for World Models via Prior-Guided Simulation

Jianan Wang, Haoquan Zhai, Siyang Zhang +7

World models learn internal representations of environment dynamics to predict future states, enabling agents to optimize action plans without physical interactions. However, devel…