rational-agents.html (3272B)
1 <!DOCTYPE html> 2 <html> 3 <head> 4 <script type="text/javascript" async src="https://cdn.jsdelivr.net/gh/mathjax/MathJax@2.7.5/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script> 5 <link rel="Stylesheet" type="text/css" href="style.css"> 6 <title>rational-agents</title> 7 <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> 8 </head> 9 <body> 10 11 <div id="Rational agents"><h1 id="Rational agents">Rational agents</h1></div> 12 13 <p> 14 "A rational agent chooses whichever action maximizes the expected value of the performance measure given the percept sequence to date and prior environment knowledge." 15 </p> 16 17 <div id="Rational agents-Agents"><h2 id="Agents">Agents</h2></div> 18 <p> 19 agent function maps percept sequence to actions (\(f: P* \rightarrow A\)) 20 </p> 21 22 <p> 23 function is internally represented by agent program 24 </p> 25 26 <p> 27 program runs on physical architecture to produce f 28 </p> 29 30 <div id="Rational agents-Rationality"><h2 id="Rationality">Rationality</h2></div> 31 <p> 32 what is rational at a specific time depends on: 33 </p> 34 <ul> 35 <li> 36 expected value of performance measure -- heuristics 37 38 <li> 39 actions and choices -- search 40 41 <li> 42 percept sequence to date -- learning 43 44 <li> 45 prior environment-- KR 46 47 </ul> 48 49 <p> 50 rationality is not omniscience or perfection 51 </p> 52 53 <div id="Rational agents-Task environments"><h2 id="Task environments">Task environments</h2></div> 54 55 <p> 56 to design rational agent, we must specify environment (PEAS): 57 </p> 58 <ul> 59 <li> 60 performance: safety, destination, profits, legality, comfort 61 62 <li> 63 environment: streets, traffic, pedestrians, weather 64 65 <li> 66 actuators: steering, accelerating, brake, horn, speaker/display 67 68 <li> 69 sensors: video, sonar, speedometer, etc. 70 71 </ul> 72 73 <p> 74 environment types: 75 </p> 76 <ul> 77 <li> 78 observable: fully (can detect all relevant aspects with sensors) or partially 79 80 <li> 81 deterministic: (yes or no) 82 83 <li> 84 static: (yes, no, semi) 85 86 <li> 87 discrete: (yes or no) 88 89 <li> 90 single-agent: (yes or no) 91 92 </ul> 93 94 <p> 95 <img src="img/environment-types.png" alt="Environment types table" /> 96 </p> 97 98 <p> 99 For Schnapsen: 100 </p> 101 <ul> 102 <li> 103 observable: not fully 104 105 <li> 106 deterministic: yes 107 108 <li> 109 static: yes 110 111 <li> 112 discrete: yes 113 114 <li> 115 single-agent: no 116 117 </ul> 118 119 <div id="Rational agents-Agent types"><h2 id="Agent types">Agent types</h2></div> 120 121 <div id="Rational agents-Agent types-Simple Reflex"><h3 id="Simple Reflex">Simple Reflex</h3></div> 122 <p> 123 select action on basis of <em>only the current percept</em> 124 </p> 125 126 <p> 127 large reduction in possible percept/action situations 128 </p> 129 130 <p> 131 implemented using condition-action rules 132 </p> 133 134 <p> 135 only works if environment is fully observable, otherwise may result in infinite loops. 136 </p> 137 138 <div id="Rational agents-Agent types-Reflex & State"><h3 id="Reflex & State">Reflex & State</h3></div> 139 <p> 140 to tackle partially observable environments, maintain internal state 141 </p> 142 143 <p> 144 over time, update state using world knowledge. 145 </p> 146 147 <div id="Rational agents-Agent types-Goal-Based"><h3 id="Goal-Based">Goal-Based</h3></div> 148 <p> 149 agent needs a goal to know the desirable situations 150 </p> 151 152 <p> 153 future is taken into account 154 </p> 155 156 <div id="Rational agents-Agent types-Learning"><h3 id="Learning">Learning</h3></div> 157 <p> 158 teach agents instead of instructing them 159 </p> 160 161 <p> 162 very robust toward initially unknown environments. 163 </p> 164 165 </body> 166 </html>