What is embodied AI?
Embodied AI studies intelligent systems that perceive and act through a body in an environment. In robotics, the body is part of the learning and control loop rather than merely an output device.
From prediction to interaction
Traditional AI can operate entirely on digital inputs and outputs. An embodied system must contend with geometry, dynamics, uncertainty, latency, contact and the consequences of its actions.
Perceive
Gather information through cameras, depth sensors, tactile sensing, proprioception and other modalities.
Reason & plan
Interpret the environment, understand goals and select actions or subgoals under physical constraints.
Act & learn
Execute movements through actuators and use the resulting interaction as feedback for control or learning.
Embodiment is more than having a robot body
A useful distinction is between an AI model controlling a machine and a system whose intelligence is shaped by interaction with the physical world. Research approaches vary widely in how much perception, reasoning, planning and control are learned jointly.
Related topics
Robot foundation models → Research → Robots →
Editorial guide · Updated September 2026