Sense-Plan-Act

The classic control loop where a robot senses, models, plans its next move, then acts, and repeats.

This is the classic three-step process in traditional robotics: use sensors to perceive the environment and update an internal world model, plan the next action on top of that model, then hand it off to the actuators to execute, and repeat. A landmark example is Shakey the Robot at Stanford Research Institute (SRI) in the 1960s–70s, which used the STRIPS planner and A* search to break an instruction down into executable steps. The drawback is that it depends on an accurate world model, is computationally slow, and breaks down easily when the environment changes. In the mid-1980s, Rodney Brooks proposed the subsumption architecture, which builds no global model and lets perception drive behavior directly; hybrid architectures combining both approaches appeared later. Today's modular robotics pipelines still follow this basic idea, while end-to-end VLA models instead use a single network to map observations directly to actions.

ExampleGiven the instruction 'push the block off the platform,' Shakey first identifies the platform and a ramp, then plans the steps 'push the ramp into place, climb onto the platform, push the block,' and finally executes them one by one.

Also called

SPA, Hierarchical Paradigm

Related

Perception-Action Loop · Subsumption Architecture · Hierarchical Architecture · End-to-End · Task Planning · World Model

Sources

Robotic paradigm - Wikipedia
Shakey the robot - Wikipedia
Subsumption architecture - Wikipedia