Perception-Action Loop

Perception drives action, and action changes what is perceived next, in a continuous back-and-forth cycle.

The perception-action loop describes the ongoing back-and-forth between an agent and its environment: sensors produce an observation, the policy produces an action, the action changes the environment and the agent's own position, and so what is seen the next moment changes too. The idea comes from cognitive science and ecological psychology; James Gibson, Francisco Varela, and others all emphasized that perception is not passive reception but is coupled to bodily movement, and roboticist Rodney Brooks likewise argued that intelligence must connect to the world through a body. This loop is a key distinction between embodied AI and disembodied AI: an image classifier looks at one picture, gives one answer, and is done, while a robot policy must repeatedly observe, act, and observe again, often tens of times per second, to correct errors and respond to change.

ExampleAs a robot arm grasps a cup, its wrist camera sees the relative position of the cup and gripper in every frame, and the policy corrects its next move accordingly — if the cup gets bumped out of place, it can realign and follow it.

Related

Embodied AI · Closed-loop Control · Open-loop Control · Embodied Cognition · Sense-Plan-Act · Active Perception

Sources

Embodied cognition - Wikipedia
Sensory-motor coupling - Wikipedia