rational-agents.wiki (2041B)
1 = Rational agents = 2 3 "A rational agent chooses whichever action maximizes the expected value of the performance measure given the percept sequence to date and prior environment knowledge." 4 5 == Agents == 6 agent function maps percept sequence to actions ($f: P* \rightarrow A$) 7 8 function is internally represented by agent program 9 10 program runs on physical architecture to produce f 11 12 == Rationality == 13 what is rational at a specific time depends on: 14 * expected value of performance measure -- heuristics 15 * actions and choices -- search 16 * percept sequence to date -- learning 17 * prior environment-- KR 18 19 rationality is not omniscience or perfection 20 21 == Task environments == 22 23 to design rational agent, we must specify environment (PEAS): 24 * performance: safety, destination, profits, legality, comfort 25 * environment: streets, traffic, pedestrians, weather 26 * actuators: steering, accelerating, brake, horn, speaker/display 27 * sensors: video, sonar, speedometer, etc. 28 29 environment types: 30 * observable: fully (can detect all relevant aspects with sensors) or partially 31 * deterministic: (yes or no) 32 * static: (yes, no, semi) 33 * discrete: (yes or no) 34 * single-agent: (yes or no) 35 36 {{file:img/environment-types.png|Environment types table}} 37 38 For Schnapsen: 39 * observable: not fully 40 * deterministic: yes 41 * static: yes 42 * discrete: yes 43 * single-agent: no 44 45 == Agent types == 46 47 === Simple Reflex === 48 select action on basis of _only the current percept_ 49 50 large reduction in possible percept/action situations 51 52 implemented using condition-action rules 53 54 only works if environment is fully observable, otherwise may result in infinite loops. 55 56 === Reflex & State === 57 to tackle partially observable environments, maintain internal state 58 59 over time, update state using world knowledge. 60 61 === Goal-Based === 62 agent needs a goal to know the desirable situations 63 64 future is taken into account 65 66 === Learning === 67 teach agents instead of instructing them 68 69 very robust toward initially unknown environments. 70