Robots Atlas>ROBOTS ATLAS
Robotics

Perception–Plan–Control Loop

ActivePublished: 29 September 2026Updated: 29 September 2026Published
Key innovation
The canonical robot loop: perceive the environment, plan an action, execute it via control — repeated at the rate the task requires.
Category
Robotics
Abstraction level
Pattern
Operation level
Robot controlSystemInference
Use cases
Mobile navigation and autonomous drivingManipulation and graspingHumanoid locomotionEmbodied-AI / VLA systemsIndustrial arm control

How it works

Each iteration: (1) perception fuses sensor signals into a state estimate; (2) a planner (classical or a learned policy) selects an action/trajectory against the goal and constraints; (3) a controller (e.g. PID, MPC, impedance control) generates actuator commands and measures the effect. The result closes the loop back to perception. Layers are often hierarchical and asynchronous.

Problem solved

A robot in a changing world cannot blindly execute a one-off plan — it needs a continuous cycle of sensing state, replanning and correcting control.

Components

PerceptionInput

Sensor fusion and state/world estimation.

PlanningDecision

Computing an action or trajectory achieving the goal.

ControlExecution

Turning the plan into actuator commands with feedback.