The next wave of humanoid robot development is focusing on physical agents, with the core challenge being robots that not only see but also understand and act successfully. The critical vision-language-action (VLA) multimodal model competition is heating up, attracting global giants like Google, Nvidia, and OpenAI.