11. Agents & Environments
Agents & Environments
What is an Agent?
An agent is anything that perceives its environment through sensors and acts upon that environment through actuators. This is a core concept in AI — all AI systems can be viewed as agents interacting with environments.
Figure: The agent-environment loop — agents perceive, reason, and act.
PEAS Framework
PEAS describes the design of an intelligent agent:
- Performance — Success criteria for the agent
- Environment — The context in which the agent operates
- Actuators — Means by which the agent acts
- Sensors — Means by which the agent perceives
Example for a self-driving car: Performance = safety, speed, comfort; Environment = roads, traffic, weather; Actuators = steering, brakes, accelerator; Sensors = cameras, lidar, GPS.
Types of Agents
- Simple Reflex Agent — Acts based solely on current percept (e.g., thermostat)
- Model-Based Reflex Agent — Maintains internal state to track unobserved aspects
- Goal-Based Agent — Chooses actions to achieve goals, uses search and planning
- Utility-Based Agent — Maximizes a utility function, handles trade-offs
- Learning Agent — Improves performance over time through experience
Environment Types
| Type | Description | Example |
|---|---|---|
| Fully Observable | Agent sees complete state | Chess board |
| Partially Observable | Agent sees partial state | Poker (hidden cards) |
| Deterministic | Next state fully determined by action | Maze |
| Stochastic | Uncertainty in state transitions | Dice game |
| Episodic | Each action is independent | Image classification |
| Sequential | Current action affects future | Chess |
| Static | World doesn't change while agent thinks | Sudoku |
| Dynamic | World changes in real-time | Self-driving |
Practice Task: For a robot vacuum cleaner, define the PEAS framework. Identify the type of agent it uses and the type of environment it operates in. Design a simple reflex agent algorithm for the robot in pseudocode. How would you upgrade it to a model-based agent?