Don't Confuse Video Generation with Planning: The Engineering Gap Between Them

Comparing it to autonomous driving and robot planning, some people treat video generation as an omniscient planner. That's like using a weather forecast as a travel guide—the forecast tells you it will rain tomorrow, but doesn't tell you whether to bring an umbrella or take a taxi. I think Xu Danfei puncturing this bubble at RSS 2026 highlights the issue most deserving of attention in the entire embodied intelligence field. From an engineering deployment perspective, this directly relates to whether we can turn demos from papers into actually usable systems.