《人工智能02智能体.pptx》由会员分享，可在线阅读，更多相关《人工智能02智能体.pptx（24页珍藏版）》请在课桌文档上搜索。
1、第二章 智能体,教材 Artificial Intelligence A modern approach(2013)S.Russell and P.Norvig人工智能一种现代方法(第二版)学校发的教材参考概率推理、机器学习部分,理性的行动：Maximize YourExpected Utility使目标期望值最大化！,智能体,An agent is anything that can be viewed as perceivingits environment through sensorsand acting upon that environment through actuators。
2、一个智能体是能通过传感器来感知周围环境并通过执行器来行动的任何东西。Agents include humans,robots,softbots（软件机器人）,thermostats（自动调温器）,etc.Human agent人类智能体:eyes,ears,and other organs for sensors;hands,legs,mouth,and other body parts for actuatorsRobotic agent机器人智能体:cameras and infrared range finders（红外测距仪）for sensors;various motors(马达)
3、for actuators,智能体和环境,The agent function（智能体函数）maps from percept histories（感知序列）to actions（行动）:f:P*AThe agent program（智能体程序）runs on the physical architecture toproduce fagent=architecture（体系结构）+program（程序）,两地点真空吸尘器世界,Percepts:location and contents,e.g.,A,DirtyActions:Left,Right,Suck,NoOp,真空吸尘智能体,What
4、 is the right function?Can it be implemented（应用）in a small agent program?,理性智能体,An agent should strive to“do the right thing”,based on what itcan perceive and the actions it can perform.The right action isthe one that will cause the agent to be most successful对的行动就是使智能体更加成功的行动Performance measure（性能度
5、量）:An objective criterion（客观标准）for success of an agents behaviorE.g.,performance measure of a vacuum-cleaner agent couldbe：灰尘清理量耗电量噪音产生量,etc.,理性智能体,Rational Agent:For each possible percept sequence,arational agent should select an action that is expectedto maximize its performance measure,given thee
6、vidence provided by the percept sequence andwhatever built-in knowledge the agent has.理性智能体：对于每一个可能的感知序列，理性智能体应该基于已知的感知序列提供的信息，和智能体已有的先验知识，选择能够使它的性能度量最大化的行为。,理性智能体,理性与全知(all-knowing with infinite knowledge)是截然不同的Agents can perform actions in order to modify futurepercepts so as to obtain useful info
7、rmation(information gathering,exploration)智能体会为了获取有用信息而采取行动信息收集An agent is autonomous if its behavior is determinedby its own experience(with ability to learn andadapt)理性智能体应该能够自主地学习，以弥补不全面或不正确的先验知识。,理性智能体,A rational agent chooses whichever action maximizes the expected value of the performance meas
8、ure given the percept sequence to dateRational omniscient(全知的)percepts may not supply all relevant informationRational clairvoyant(洞察力)action outcomes may not be as expectedHence,rational successfulRational exploration探索,learning学习,autonomy自主性,Agent：自动驾驶汽车,Performance measure性能度量：safety,destination,
9、profits,legality,comfort,Environment环境：streets/freeways,traffic,pedestrians,weather,Actuators执行器：steering,accelerator,brake,horn,speaker/display,Sensors传感器：video,accelerometers,gauges,engine sensors,keyboard,GPS,任务环境PEAS,任务环境属性,完全可观察(vs.部分可观察):如果一直智能体的传感器每个时间节点上都能让它访问获取环境的完整状态确定性的(vs.随机的):如果环境的下一个状态
10、完全决定于当前的状态和智能体执行的动作.(如果环境是确定性的，除非有其它智能体活动的影响，那么我们称该环境是策略的)片段式的(vs.延续式的):智能体的经验被分成了一个个原子片段(每个片段的组成包含了智能体所感知的信息以及进而执行的单个行动),and 行动的选择只取决于当前片段自身.如装配线上检测次品零件的机器人只需要把每次决策建立在当前零件基础上，不用考虑以前的决策。,任务环境属性,静态的(vs.动态的):如果环境在智能体思考的时候不会变化.(如果环境本身不随时间的流逝变化，但智能体的性能评价随时间变化，那么我们称这个环境是半动态的)离散的(vs.连续的):包含有限个数的独特状态.下棋、开车
11、单智能体(vs.多智能体):An agent operating byitself in an environment.,任务环境属性,任务环境属性取决于任务环境是如何定义的The real world is(of course)部分可观察的,随机的,延续式的,动态的,连续的,多智能体的,智能体函数和程序,An agent is completely specified by the agentfunction mapping percept sequences to actionsOne agent function(or a small equivalence class)is ratio
12、nal目标:find a way to implement the rational agentfunction concisely,Table-lookup agent,inputalgorithms/table-agent-algorithm缺陷:空间大创建表的时间太长没有自主性即使采用学习的方法，智能体在有限时间 内无法学习所有表项,智能体类型,四种基本的智能体程序，按通用性递增排序:simple reflex agents简单反射型智能体 reflex agents with state基于模型的反射型智能体 goal-based agents基于目标的智能体 utility-base
13、d agents基于效用的智能体All these can be turned into learning agents学习智能体,简单反射型智能体,基于模型的反射型智能体,基于目标的智能体,基于效用的智能体,学习智能体,Summary,Agents interact with environments through actuators and sensorsThe agent function describes what the agent does in all circumstancesThe performance measure evaluates the environment
14、 sequenceA perfectly rational agent maximizes expected performanceAgent programs implement(some)agent functionsPEAS descriptions define task environmentsEnvironments are categorized along several dimensions:observable?deterministic?episodic?static?discrete?single-agent?Several basic agent architectures exist:reflex,reflex with state,goal-based,utility-based,
