Embodied AGI

Embodied intelligence able to perform diverse open-ended real-world tasks the way a human can — the field's long-term goal.

Embodied AGI is a goal concept combining artificial general intelligence (AGI) with embodied AI; there is currently no single agreed-upon standard for it. A 2025 survey by Fan Wang and Hao Sun offers a working definition: embodied intelligence that shows human-like interaction ability and can perform diverse, open-ended, real-world tasks at a human level. The paper proposes an L1–L5 grading scale modeled on autonomous-driving levels: L1 completes a single task, L2 completes a combination of tasks, L3 conditionally completes general tasks, and L4–L5 progressively move toward open-ended tasks, human-like behavior, and no need for human intervention. It evaluates systems along four dimensions — omni-modality, human-like cognition, real-time responsiveness, and generalization — and judges that embodied AI today sits roughly between L1 and L2. Grading schemes like this are mainly used to describe direction and measure progress, not as a strict technical standard.

ExampleUnder this scale, a robot arm that can only pick up one type of part on a production line is L1; a robot that can break down “make a cup of coffee” into preset skills such as getting a cup, adding water, and pressing buttons, and complete them in order, is close to L2.

Also called

General Embodied Intelligence

Related

Embodied AI · General-purpose Robot · Physical Turing Test · Levels of Autonomy · Humanoid Robot Intelligence Level Grading

Sources

Toward Embodied AGI: A Review of Embodied AI and the Road Ahead (arXiv 2505.14235)

As of

2025-05