Artificial Intelligence: Task Environments

Study Notes

Task environments represent problem spaces requiring agent solutions and directly influence agent design.
Task environments are specified using the PEAS framework:
- Performance Measure
- Environment
- Actuators
- Sensors

Autonomous Taxi
- Performance Measure: Safe, fast, legal, and comfortable trips.
- Environment: Roads, traffic, pedestrians, and customers.
- Actuators: Steering, accelerator, brake, signal, horn, and display.
- Sensors: Cameras, sonar, speedometer, GPS, accelerometer, and engine sensors.
Part Picking Robot
- Performance Measure: Percentage of parts placed in correct bins.
- Environment: Conveyor belt with parts and bins.
- Actuators: Jointed arm and hand.
- Sensors: Camera and joint angle sensor.

Fully Observable vs. Partially Observable
- Fully Observable: Sensors detect all aspects relevant to actions; agents do not maintain internal state.
- Partially Observable: Inaccurate/noisy sensors lead to missing information; agents cannot assess all states (e.g., vacuum cleaner's limited view).
Unobservable
- Agents lack any sensors to perceive the environment.

Single Agent
- Example: Solving a crossword puzzle.
Multiple Agent
- Example: Playing chess where two agents interact to optimize their performance measures.
Competitive Multi-Agent Environment
- Example: Chess involves maximizing one's performance while minimizing the opponent's.
Cooperative Multi-Agent Environment
- Example: Taxi driving where agents help each other avoid collisions.

Deterministic vs. Stochastic
- Deterministic: Future state completely determined by current state and actions (e.g., vacuum world).
- Non-Deterministic: Uncertainty exists due to incomplete information or variation in outcomes (e.g., taxi driving).
- Stochastic: Uncertainty characterized by probabilities of possible outcomes.
Episodic vs. Sequential
- Episodic: Experience divided into episodes; current decision independent of past actions (e.g., classifying parts).
- Sequential: Decisions impact future choices (e.g., chess, taxi driving).

Static vs. Dynamic
- Static: Environment doesn't change during agent deliberation (e.g., crossword puzzle).
- Dynamic: Environment state can change while the agent acts (e.g., autonomous taxi driving).
Semi-Dynamic
- Environment does not change, but agent performance score alters over time (e.g., chess with a clock).

Discrete vs. Continuous
- Discrete: Finite distinct states and actions (e.g., chess).
- Continuous: Infinite possible states and actions (e.g., taxi driving).
Known vs. Unknown
- Known: Outcomes or probabilities for all actions are provided; can be partially observable (e.g., Solitaire).
- Unknown: Agent learns about the environment to make effective decisions; can be fully observable.

The most challenging scenario combines:
- Partially observable, multi-agent, stochastic, sequential, dynamic, continuous, and unknown conditions (e.g., taxi driving).