An embodied intelligence system combining causal world models and vision-language-action models to transfer skills across different robot forms.