Spatial state and embodied world models

WorldBridge, HiFOR, CausalWAM, and CaPtain extend work on video-to-action transfer, action-guided foresight, causal action representations, and code-as-policy robotics. Their reported simulation and robot results are promising, but transfer, baselines, and evaluation conditions need scrutiny; earlier work includes FrameForge, Wizard, and ReLift.