PEAS Description — Beginner-Friendly Notes
1. What PEAS Means
When you design an agent, describe the task with PEAS:
- P — Performance measure: how success is scored
- E — Environment: where the agent operates
- A — Actuators: what the agent can do to the environment
- S — Sensors: what the agent can perceive
No PEAS → no clear task → cannot build or test the agent.
2. How to Use It — 3 Worked Examples
Example A: Self-Driving Taxi
- P: safety, time to destination, comfort, fuel, obey traffic rules, profit
- E: roads, traffic, pedestrians, weather, passengers
- A: steering, accelerator, brake, horn, display/speaker
- S: cameras, lidar, GPS, speedometer, microphone
Example B: Medical Diagnosis Support (AI doctor assistant)
- P: accuracy, cost, patient survival, minimal false negatives
- E: patients, hospital, records, other doctors
- A: show diagnosis, order tests, schedule appointments
- S: keyboard/mouse, patient records, test results, camera
Example C: Vacuum Robot (exam favourite)
- P: cleanliness, efficiency, battery life, noise
- E: rooms, dirt, obstacles, floor type
- A: wheels, suction, brush
- S: camera, dirt sensor, bump sensor, battery sensor
3. Mini-Template for Any Question
Task: ___________
P: ___________ (what to maximise/minimise)
E: ___________ (everything the agent cannot control)
A: ___________ (actions available)
S: ___________ (inputs available)
Fill in that order — examiners look for completeness.
4. Common Mistakes
- Listing an actuator as a sensor (e.g., wheels are actuators, not sensors).
- Forgetting P is measurable — "do well" is not a performance measure; "percentage of dirt cleaned in 30 min" is.
- Mixing environment with actuators — environment is external world; actuators are the agent's means to change it.
5. PEAS Leads to Design Choices
PEAS tells you how hard the problem is: if E is observable, deterministic, static, discrete, single-agent → easy; if opposite → needs learning and complex agents (next topics).
6. One-Line Summary to Write in Exam
PEAS specifies an agent task completely as Performance, Environment, Actuators and Sensors, making it possible to design and evaluate a rational agent.
7. Quick Self-Check
- Write PEAS for a chess-playing agent.
- Write PEAS for an online shopping recommendation agent.
- Which part of PEAS would you change if you add "must explain its diagnosis" to a medical agent?