2.4 The PEAS Framework (BT104CO)

1. What is PEAS?

In the framework of Russell and Norvig, PEAS is the first step in designing any AI. Before writing code, you must specify the task environment using these four headings.

P

Performance

The objective criteria for success. What the agent is trying to maximize (the "goal" or "reward").

E

Environment

The external world or setting where the agent operates (roads, patient, image database).

A

Actuators

The "tools" the agent uses to perform actions and change its environment (steering, display, tags).

S

Sensors

The inputs the agent uses to perceive the environment and gather data (cameras, keyboard, LIDAR).

2. PEAS Examples (Case Studies)

To master PEAS for an exam, you should be able to break down any AI system into these four categories.

Case A: Automated Taxi Driver

CategoryDescription
PerformanceSafety, speed, legal driving, passenger comfort, profit.
EnvironmentRoads, other traffic, pedestrians, weather, customers.
ActuatorsSteering wheel, accelerator, brake, signal, horn, display.
SensorsCameras, LIDAR, GPS, speedometer, engine sensors.

Case B: Medical Diagnosis System

CategoryDescription
PerformanceHealthy patient, minimized costs, avoiding malpractice.
EnvironmentPatient, hospital staff, medical database.
ActuatorsDisplay of questions, tests, diagnoses, treatments.
SensorsKeyboard/Touchscreen (symptoms, lab results, history).

Case C: Satellite Image Analysis

CategoryDescription
PerformanceCategorization accuracy, speed, minimized false positives.
EnvironmentDownlink from satellite, image database, weather.
ActuatorsTagging images, sending alerts, re-ordering database.
SensorsHigh-resolution cameras, infrared sensors.

3. The Relationship Between PEAS and Rationality

A Rational Agent is defined relative to its PEAS. An action is rational if it follows this logic:

1. Maximize the Performance measure...
2. Given the history of Sensors (percepts)...
3. Within the specified Environment...
4. Using the available Actuators.

The "Vacuum Cleaner" Gotcha

Performance Measure Selection: If you give a vacuum cleaner a measure of "amount of dust collected," it might suck up dust, spit it out, and suck it up again to "maximize" its score. This is irrational. A better measure is "Cleanliness of the floor."

Practice Quiz