Humanoid AI with Humanity Modeling: Bridging Humans and Virtual–Physical Humanoids

The seventy-year AI is now entering another major resurgence, the era of Generative AI (GenAI), characterized by the rapid advancement of large multimodal models (LMMs). GenAI and LMMs are fundamentally transforming sixty-decade-old humanoid robots, advancing them from human-looking and humanlike embodiments toward human-level AI and robotics, and inspiring a new era of AI-robotics symbiosis: humanoid AI . Humanoid AI develops humane AI-driven humanoids (AI humanoids) that go beyond traditional human-looking robots, which typically exhibit little or no humanity and intelligence due to their dedication to predefined and limited AI functions and exteriors. By contrast, AI humanoids undertake real-time, situated, and interactive mind-to-action, perception-to-behavior, and vision-to-emotion transformations, achieving an unprecedented degree of replicating human characteristics and intelligence. Realizing this vision requires systematic humanity modeling, ensuring that AI humanoids are shaped, guided, and governed by robust frameworks of AI and robotic humanity. We then propose human-virtual-physical humanoid transform and alignment and human-metaverse-humanoid ecosystem for open, complex human-digital-machine-cooperative methodology and problem-solving, followed by general principles and frameworks for building actionable humanoids and evaluating humanoid quality.