Physical AGI

Physical AGI is Xu Huazhe’s phrase in 166: 许华哲再次具身创业：不想错过最大的西瓜 for general intelligence that can act in the physical world. It is narrower than all Physical AI because it implies broad task generality, but broader than a single robot product because the key asset is the general robot brain rather than the shell.

The source contrasts Physical AGI with two weaker targets: making a robot look more human, and making a specialized robot work well in one industrial scene. Xu argues that the “largest watermelon” is the intelligence layer that can generalize across homes, objects, tasks, and changing physical situations.

Key Claims

• A human-like body is not enough; the system has to complete different tasks across different scenes.
• Industrial and logistics scenarios can create useful near-term businesses, but may not force the same generality as a household environment.
• World Models, Vision Language Action Models, reinforcement learning, video data, and real robot data may all contribute, but none should be treated as a finished recipe.
• Product design still matters because early Physical AGI systems should restrict high-risk actions until model reliability improves.

Connections

• Physical AI and Embodied AI — broader categories.
• Poke Robotics and Xu Huazhe — company and founder pursuing the route.
• AI Native Robotics and Unified Robot Models — route principles attached to the concept.
• Home Service Robots and Household Robot Data Flywheel — household deployment case.
• World Models, Vision Language Action Models, and World Model VLA Fusion — model-route context.