120 likes | 353 Views
Designing an Intelligent Agent. CAP 4621 (Artificial Intelligence) Fall 2002, University of Florida Eric Spellman. What this lecture covers. General architectures for agents Reflex agent Goal-based agent Utility-based agent Describing environments
E N D
Designing an Intelligent Agent CAP 4621 (Artificial Intelligence) Fall 2002, University of Florida Eric Spellman
What this lecture covers • General architectures for agents • Reflex agent • Goal-based agent • Utility-based agent • Describing environments • accessible? deterministic? episodic? static? discrete?
Choosing your architecture • Your problem dictates your agent. • Typically default to simplest which works • Must your agent deal with surprises? • How complex must his behavior be?
Reflex agents • Not autonomous, must specify all behavior • Many, many condition-action rules • Suited for a strictly controlled environment • Suited for repetitive tasks (episodic) • Example: Airline passenger screening
Adding a memory • Many problems require a series of steps to succeed. • Agents for such problems must have “internal state” (a memory). • Actions depend on precepts and state. • What do you need to store? (Memento) • Examples: Target tracking; Shell game.
Goal-based agents • Much easier to specify behavior • Tell agent what to do not how to do it. • Agent predicts how his actions effect world. • Uses search techniques to find sequence of actions leading to a goal state. • More flexible than reflex. • Example: Simple game playing.
Utility-based agent • Quantify desirability of states • Can select among conflicting goals • Can act even when it cannot predict the exact sequence leading to a goal. • More powerful than goal-based agents. • Examples: Chess (discrete); Moving a robot arm (continuous).
Accessible Ex: Chess Deterministic Ex: Poker (w/ cheating) Episodic Ex: Passenger profiling Inaccessible Ex: Shell game Nondeterministic Ex: Poker Nonepisodic Ex: Profiling for groups Describing environments
Static Ex: Chess Dynamic Ex: Target tracking Describing environments (cont.) Semidynamic Ex: Credit card authorization Discrete Ex: Chess Continuous Ex: Target tracking
Conclusion • Architectures of different complexities • Must describe and understand your problem before choosing which architecture. • No free lunch: • Simple architecture and complex rules set • Flexibility requires complex architecture.