Embodied AI

Artificial intelligence with a body, able to perceive, decide, and act in the physical world.

Embodied AI refers to intelligent systems that have a physical body (or a simulated one) and perceive, decide, and act through real-time interaction with their environment — robots and self-driving cars are the typical examples. The contrast is disembodied AI, like ChatGPT, which only processes text and images and never directly acts on the physical world. The idea is often traced back to Alan Turing's 1950 paper “Computing Machinery and Intelligence,” and to the robotics research of Rodney Brooks and others in the 1980s–90s, which emphasized that a body interacting with its environment is central to intelligence. At its core is a closed loop of perception, decision, action, and feedback. In the past few years, large language models and vision-language models have let robots understand open-ended instructions, making embodied AI a hot field, often seen as a key step toward artificial general intelligence (AGI).

ExampleA home robot told to put the dirty clothes in the washing machine has to find the clothes itself, walk over, pick them up, open the machine, and load them, adjusting its actions throughout using visual feedback — that is embodied AI. ChatGPT can only describe the steps; that is disembodied AI.

Also called

Embodied Intelligence, EAI, Embodied Artificial Intelligence

Related

Disembodied AI · Physical AI · Embodied Agent · Embodied Cognition · General-purpose Robot · Vision-Language-Action Model

Sources

维基百科：具身智能 (Chinese)
Aligning Cyber Space with Physical World: A Comprehensive Survey on Embodied AI