Download Agents - Introduction to Artificial Intelligence | CS 440 and more Study notes Computer Science in PDF only on Docsity! 1 Agents Lecture #2 8/28/08 What is an agent? According to your text: “An agent is anything that can be viewed as perceiving its environment through sensors and acting on that environment through actuators.” (p. 32) Caveat Your book also says: “the notion of an agent is meant to be a tool for analyzing systems, not an absolute characterization that divides the world into agents and non-agents. One could view a hand-held calculator as an agent that chooses the action of displaying “4” when given the percept sequence “2+2= ”, but such an analysis would hardly aid our understanding of the calculator.” (p. 34, emphasis mine) Examples of agents Taxi driver Internet shopper Backgammon player Chemical plant controller Spam detector Agents E nvironm ent ? sensors Actuators (my redrawing of Figure 2.1 in your text) Actions Percepts1 1) This is a poor choice of words on Russell & Norvig’s part. It would be more consistent with the psychology & philosophy literature to say that the environment provided stimuli, and percepts are the agent’s interpretations of the stimuli. But we will use R&N’s terminology. The agent and the environment An agent: Works in a particular environment Has goals Perceives the environment Performs actions to achieve its goals. 2 Formalizing Task Environments (PEAS) P: Performance This is all important: it defines the goal E: Environment This defines the world the agent lives in A: Actuators This defines how the agent can change the world S: Sensors This defines how the agent sees the world (and how much of the word the agent can see) Example: the automated taxi driver Environment: – Roads, other traffic, pedestrians, other impediments Sensors: – Cameras, engine sensors, laser range finders, GPS, keyboard, microphone Actuators: – Steering, accelerator, brake, turn signal, horn. Possible Performance Measures: – Safety, speed, legality, comfort, profit. Formalizing Agents Agent functions map sequences of percepts onto actions APf →*: Formalizing Agents (II) Really? Do you believe the last slide? How about… ( ) AAPf →**,: Environments Discrete Static Agents Episodic Deterministic partially?fullyfullyObservable TaxiInternet shopping BackgammonCrossword puzzle Fully vs. partially observable: can the sensors detect all aspects that are relevant to the choice of action. Environments Discrete Static Agents Episodic stochastic?stochasticdeterministicDeterministic partially?fullyfullyObservable TaxiInternet shopping BackgammonCrossword puzzle Deterministic vs. stochastic: is the next environment state completely determined by the current state?