Embodied Agent

An agent with a physical or virtual body that perceives and acts within an environment through that body.

An agent is a system that can perceive its environment and take action to reach a goal. An embodied agent emphasizes that it has a body, and perceives and acts in the environment through that body. Wikipedia defines it as an agent that interacts with its environment through a physical body situated within that environment; more broadly, this also includes virtual bodies in a simulator or a game. It differs from a purely conversational agent that only outputs text, since it needs to chain together instruction understanding, active exploration, multimodal perception, and executed action into a working loop. The 2024 survey by Yang Liu and colleagues lists embodied agents as one of embodied AI's four main research goals, and argues that multimodal large models are currently its dominant “brain.” Real robots, self-driving cars, and navigation agents in simulators such as Habitat all count as embodied agents.

ExampleIn the Habitat home simulator, a virtual robot with a camera is told to go find a cup in the kitchen; it moves and looks around on its own and eventually stops in front of the cup — that makes it an embodied agent.

Related

Embodied AI · Agent–Environment Interaction · Perception-Action Loop · Disembodied AI · Policy

Sources

Embodied agent - Wikipedia
Aligning Cyber Space with Physical World: A Comprehensive Survey on Embodied AI (arXiv 2407.06886)