1、1/33 Agent Agent及其理性 PEAS及环境特性 Agent程序的结构 3/33 AGENT及其理性 4/33 Intelligent agent 2 Example MS agent Robot 5/33 Agents 2.1 An agent is anything that can be viewed as perceiving(感知 ) its environment through sensors(传感器 ) acting(动作 ) upon that environment through actuators(执行器 ) humans, robots, softbots
2、, 7/33 the structure of a general agent 8/33 Agents and Environments 2.1 An agent program is running on the (physical) architecture to produce the agent function. The agent function maps from percept histories to actions: The agents choice of action at any given instant can depend on the entire perc
3、ept sequence observed to date. EnvironmentAgent Sensors Actuators ? Percepts Actions 9/33 Vacuum-cleaner agent 2.1 Percepts: location and contents, e.g. A, dirty (Idealization: locations are discrete) Actions: LEFT, RIGHT, SUCK A B 10/33 A Reflex Vacuum-Cleaner 2.1 A, Clean A, Dirty B, Clean B, Dirt
4、y A, Clean, A, Clean A, Clean, A, Dirty Right Suck Left Suck Right Suck Percept sequence Action 填写表格右边一列的方法不同，就可以定义不同的真空吸尘器世界Agent。 11/33 rationality depends on 12/33 Good behavior: The concept of Rationality 2.2 Rationality depends on The performance measure that defines the criterion of success Th
5、e agents prior knowledge of the environment The actions that the agent can perform The agents percept sequence to date 13/33 Good behavior: The concept of Rationality 2.2.1 The definition of rational agent For each possible percept sequence, a rational agent should select an action that is expected
6、to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge the agent has. 14/33 Good behavior: The concept of Rationality 2.2.1 是否是理性的？ 每个时间步清洁一个方块加 1分，共 1000步 地形已知，灰尘分布和初始位置未知。吸尘清洁当前方格，干净的方格保持干净。 Left和Right使 Agent移动，但不出界。 行动只有 Left、 Righ
7、t、 Suck Agent能感知位置以及所在方格是否有灰尘 15/33 Good behavior: The concept of Rationality 2.2.1 是否是理性的？ 每个时间步清洁一个方块加 1分，共 1000步 地形已知， 所有方格均干净 。 Left和 Right使Agent移动，但不出界。 行动只有 Left、 Right、 Suck Agent能感知位置以及所在方格是否有灰尘 16/33 Good behavior: The concept of Rationality 2.2.2 理性 ？ 全知 17/33 Good behavior: The concept of
8、 Rationality 2.2.2 理性 全知 全知的 Agent明确知道它的动作产生的实际结果 理性 ？ 完美 18/33 Good behavior: The concept of Rationality 2.2.2 理性 全知 全知的 Agent明确知道它的动作产生的实际结果 理性 完美 理性 Agent使期望性能最大化；完美 Agent使实际性能最大化。 理性 Agent还应该： 19/33 Good behavior: The concept of Rationality 2.2.2 理性 全知 全知的 Agent明确知道它的动作产生的实际结果 理性 完美 理性 Agent使期望性
9、能最大化；完美 Agent使实际性能最大化。 理性 Agent还应该 收集信息、学习 22/33 PEAS及环境特性 23/33 The nature of environments 2.3.1 To design a rational agent we need to specify a task environment PEAS: to specify a task environment ？ ？ ？ ？ 24/33 The nature of environments 2.3.1 To design a rational agent we need to specify a task e
10、nvironment PEAS: to specify a task environment Performance measure Environment Actuators Sensors 25/33 PEAS: autobot 2.3.1 Specifying an automated taxi driver Performance measure ？ Environment ？ Actuators ？ Sensors ？ 26/33 PEAS: autobot 2.3.1 Specifying an automated taxi driver Performance measure s
11、afe, fast, legal, comfortable, maximize profits Environment roads, other traffic, pedestrians(行人 ), customers Actuators Steering(方向盘 ), accelerator(加速器 ), brake(刹车 ), signal, horn(喇叭 ) Sensors cameras, sonar(声纳 ), speedometer(速度计 ), GPS 27/33 PEAS Spam Filtering : Performance measure: Environment: A
12、ctuators: Sensors: 28/33 PEAS Spam Filtering : Performance measure: spam block Environment: email client or server Actuators: mark as spam, transfer messages Sensors: emails (possibly across users), etc. 29/33 Properties of task environments 2.3.2 Fully vs. partially observable can the sensors detec
13、t all aspects that are relevant to the choice of action? Deterministic vs. stochastic(随机 ) if the next environment state is completely determined by the current state and the executed action? Episodic(片断式的 ) vs. sequential(延续式的 ) if the agents experience can be divided into atomic steps and the choi
14、ce of action depends only on the episode itself? 30/33 Properties of task environments 2.3.2 Static vs. dynamic If the environment can change while the agent is choosing an action? Discrete vs. continuous This distinction can be applied to the state of the environment, the way time is handled and to
15、 the percepts/actions of the agent. Single vs. multi-agent Does the environment contain other agents who are also maximizing some performance measure that depends on the current agents actions? 31/33 32/33 Agents and environments 2.3.2 Agents Discrete Static Episodic Deterministic Observable Taxi Po
16、ker Crossword puzzle Task The environment type largely determines the agent design 33/33 Agents and environments 2.3.2 multi multi Single Agents continuous discrete discrete Discrete dynamic static static Static sequential sequential sequential Episodic stochastic stochastic deterministic Deterministic Partially partially fully Observable Taxi Poker Crossword puzzle Task The environment type largely determines the agent design 34/33 Environment Simplifications 2.3.2 The real world Partially observable Stochastic Sequential Dynamic Continuous Multi-agents
