Introduction Week 1, Lecture 1

Transcription

1 CS 485/680 Knowledge-Based Agents Introduction Week 1, Lecture 1 William Regli, Vincent Cicirello, Maxim Peysakhov, Joe Kopena Geometric and Intelligent Computing Laboratory Department of Computer Science Drexel University 1 Team Taught Course Who? Dr. William Regli Dr. Vince Cicirello Mr. Max Peysakhov TA: Joe Kopena Why? Multi-agent systems draw from many disciplines Take advantage of varied expertise and background of instructors Dr. Regli: applied AI, multi-agent systems, agent system security, network issues Dr. Cicirello: multi-agent coordination, swarm-inspired agent models, multi-agent planning/scheduling, robotics Mr. Peysakhov: emergent behavior, swarming agents, mobile agents Joe Kopena: service-based computing, knowledge bases, ontologies Course Overview What is this course all about? Official title: Knowledge-Based Agents Course is more than just about knowledge-based issues and models Think of it more as about the multi-splendored world of agent-based computing (from suggested title change of Joe Kopena) Highly interdisciplinary, drawing from diverse fields, including: AI (especially Distributed AI), economics, artificial life, game theory, social science, etc. 3 Course Outline Introduction (tonight s lecture) Part I: Simple agents (emergent behavior, swarm intelligence) weeks and 3 Part II: Representation & reasoning (logics, inference, planning, knowledge bases, etc) weeks 4 and 5 Part III: Multi-Agent Coordination (game theory, auctions, negotiation) weeks 6 and 7 Part IV: Other issues (mobility, network related issues, security) weeks 9 and 10 4 Course Management Issues All course policies are in the syllabus Extensive use of PDF handouts Must read list every day There will be 5 programming assignments and -3 problem sets (plan on minimum of 10-to-15 hrs per week) Suggestion: print out handouts before class, use them to take notes READ THE SYLLABUS!! 5 Collaboration Policies Thou Shall write your own code do your own math attribute any work that is not your own talk among yourselves, share ideas Thou Shall Not Share code Use ideas without attribution All code will be auto checked for plagiarism Violations will result in an automatic F 6 1

2 Course Policies Late policy: NO late homeworks, No excuses. (Some considerations may be given for a valid medical reason) NO Cell phones! Cell phone ring = extra homework For everyone your friends will love you NO cheating. Penalty = F. No mercy. No extra homework's, projects or tests to improve grade will be given after the final. Partial credit for implementing some of the requirements. Code that does not compile or does not produce output Required Books Required An Introduction to MultiAgent The Computational Beauty of will get 0 points. Systems, by Michael Nature, by Gary William 7 8 Wooldgridge; ISBN: Flake; ISBN: What is an agent?, a Definition Go to class web page An agent is a computer system that is capable of independent action on behalf of its user or owner (figuring out what needs to be done to satisfy design objectives, rather than constantly being told) [ From Wooldridge textbook] 9 10 What is an Agent? Another definition An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through effectors. (Russell and Norvig) Some properties of an intelligent agent: Autonomous: operates without direct intervention; some form of control over actions and internal state Reactive: reactions to external changes in environment Pro-active: goal-directed Social: interacts with others to achieve goals What is an Agent? The main point about agents is they are autonomous: capable of acting independently, exhibiting control over their internal state Thus: an agent is capable of autonomous action in some environment in order to meet its design objectives input SYSTEM output 11 ENVIRONMENT 1

3 What is an Agent? Trivial (non-interesting) agents: thermostat UNIX daemon (e.g., biff) An intelligent agent is a computer system capable of flexible autonomous action in some environment By flexible, we mean: reactive pro-active social 13 Reactivity If a program s environment is guaranteed to be fixed, the program need never worry about its own success or failure program just executes blindly Example of fixed environment: compiler The real world is not like that: things change, information is incomplete. Many (most?) interesting environments are dynamic Software is hard to build for dynamic domains: program must take into account possibility of failure ask itself whether it is worth executing! A reactive system is one that maintains an ongoing interaction with its environment, and responds to changes that occur in it (in time for the response to be useful) 14 Proactiveness Reacting to an environment is easy (e.g., stimulus response rules) But we generally want agents to do things for us Hence goal directed behavior Pro-activeness = generating and attempting to achieve goals; not driven solely by events; taking the initiative Recognizing opportunities 15 Balancing Reactive and Goal- Oriented Behavior We want our agents to be reactive, responding to changing conditions in an appropriate (timely) fashion We want our agents to systematically work towards long-term goals These two considerations can be at odds with one another Designing an agent that can balance the two remains an open research problem 16 Social Ability The real world is a multi-agent environment: we cannot go around attempting to achieve goals without taking others into account Some goals can only be achieved with the cooperation of others Similarly for many computer environments: witness the Internet Social ability in agents is the ability to interact with other agents (and possibly humans) via some kind of agent-communication language, Multiagent Systems, a Definition A multiagent system is one that consists of a number of agents, which interact with oneanother In the most general case, agents act on behalf of users with different goals and motivations To successfully interact, they require the ability to cooperate, coordinate, and negotiate with each other and perhaps cooperate with others

4 Other Properties Other properties, sometimes discussed in the context of agency: mobility: the ability of an agent to move around an electronic network veracity: an agent will not knowingly communicate false information benevolence: agents do not have conflicting goals, and that every agent will therefore always try to do what is asked of it rationality: agent will act in order to achieve its goals, and will not act in such a way as to prevent its goals being achieved at least insofar as its beliefs permit learning/adaption: agents improve performance over time physical embodiment: robots 19 Example: Microsoft Office Assistant Reactive: senses and responds to human user actions Pro-active: behavior directed towards goal of assisting human user Social: interacts with human user Veracity: provides human user with true information Benevolence Learning/adaptation 0 Environments: Accessible vs. inaccessible An accessible environment is one in which the agent can obtain complete, accurate, up-to-date information about the environment s state Most moderately complex environments (including, for example, the everyday physical world and the Internet) are inaccessible The more accessible an environment is, the simpler it is to build agents to operate in it Environments: Deterministic vs. non-deterministic A deterministic environment is one in which any action has a single guaranteed effect there is no uncertainty about the state that will result from performing an action The physical world can to all intents and purposes be regarded as non-deterministic environments present greater problems for the agent designer 1 Environments: Episodic vs. non-episodic In an episodic environment, the performance of an agent is dependent on a number of discrete episodes, with no link between the performance of an agent in different scenarios Episodic environments are simpler from the agent developer s perspective because the agent can decide what action to perform based only on the current episode it need not reason about the interactions between this and future episodes 3 Environments: Static vs. dynamic A static environment is one that can be assumed to remain unchanged except by the performance of actions by the agent A dynamic environment is one that has other processes operating on it, and which hence changes in ways beyond the agent s control Other processes can interfere with the agent s actions (as in concurrent systems theory) The physical world is a highly dynamic environment 4 4

5 Environments: Discrete vs. continuous An environment is discrete if there are a fixed, finite number of actions and percepts in it Russell and Norvig give a chess game as an example of a discrete environment, and taxi driving as an example of a continuous one Continuous environments have a certain level of mismatch with computer systems Discrete environments could in principle be handled by a kind of lookup table Agents as software engineering paradigm From Object-oriented to Agent-Oriented Are agents just objects by another name? Object: encapsulates some state communicates via message passing has methods, corresponding to operations that may be performed on this state 5 6 Agents and Objects Main differences: agents are autonomous: agents embody stronger notion of autonomy than objects, and in particular, they decide for themselves whether or not to perform an action on request from another agent agents are smart: capable of flexible (reactive, pro-active, social) behavior, and the standard object model has nothing to say about such types of behavior agents are active: a multi-agent system is inherently multi-threaded, in that each agent is assumed to have at least one thread of active control 7 8 Objects do it for free agents do it because they want to agents do it for money 9 Objects vs. Agents Objects passive encapsulates data / actions performed on that data control internal state do not control behavior cannot prevent others from using public methods perform requests for free part of a larger program with single thread of execution can be mobile or distributed Agents active an object with reasoning ability and goal-directed behavior control internal state do control behavior complete control over who and how they are used perform requests if they gain something own thread of execution can be actively mobile (choose where, when, why to migrate) 30 5

6 Applications of Agent Technology Application Domains Example Problem Domains Robotic Space Exploration Transportation Web Search Tools Interface animation Information brokers Applets Semantic broker and name space services Electronic commerce & message-passing Robots Dynamic services Control protocols Softbots K-based system Robotic Soccer Manufacturing Intelligent routers Key aspects of an agent system # of agents # of agent types Complexity (# of actions an agent could take) Current tools cluster near the axes on the orthogonal planes The Agent System Number of agent types Number of agents Agent complexity 33 Number of Agents X Example: Simple Agents Agent Complexity Termites Ants Flocking Birds all examples of emergent behavior in very simple agents. 34 Agent-Based Systems: The Major Issues Knowledge Representation Reasoning/Action Infrastructures Communication Ontologies Frameworks Architecture Agent Building Tools Knowledge Representation Implicit Knowledge recorded in the program C/C++ Perl LISP Explicit Knowledge recorded in rules Prolog Rules Knowledge Interchange Format (KIF)

7 Issues for Agents & Representation Representation is separated from algorithms/reasoning from agent-agent communication Representations usually domain specific Integration of representations is hard Problems studied by those who do Knowledge Engineering Reasoning algorithms + knowledge => actions Algorithms depend on application often needs customization/tuning for domain real time, latency, bandwidth, etc. Knowledge in common format Agent takes actions to complete task Techniques for Reasoning Issues for Agent Reasoning Rules, Logic Statements, Expert Systems State searches Planning, Scheduling Bayesian Belief Networks Neural Networks Genetic Algorithms etc. Applications Optimization Natural Language Processing (NLP) Natural Language Translation (NLT) Data mining Robotics Time and Complexity Distributed Decision Making Distributed Control assignment of roles and duties when agent network is under stress Security and Information Assurance Trust and Beliefs, Incomplete Information Agent Communication: Issues Different computer systems Different computer programs Common language is needed Directory Services Ie: where to get information Routing/Addressing Bandwidth Agent Infrastructures Regulations that the agents follow Deal with Ontologies (representation) Communication Infrastructures Transport Protocols Interaction Protocols

8 Infrastructure: Communications Examples: Blackboard Broadcast/Subscribe Peer--Peer Example: Blackboard 43 Agents share common memory space for communication Memory space does not have to be on the same computer 44 Example: Broadcast/Subscribe Fault tolerant Agents replaced easily Security is lacking Large numbers of agents affect performance Example: Peer--Peer Knows where messages are sent Easy to introduce security measures Hard to work with in open agent systems Gnutella, Kazza Infrastructure: Transport Protocols TCP/IP, SMTP, HTTP Primarily concerned with getting data from point A to point B CORBA, OLE, DCOM, and OpenDoc Concerned with exchanging procedural information and datastructures SKTP Exchanging of knowledge between agents Infrastructure: Interaction Protocols Dictate how agents talk to each other Coordination Conflict resolution Negotiation

9 Issues in Agent Infrastructures Common and Shared Representations Agile Agent Building Tools shells and agent toolkits support variety of logical agent frameworks Realism and Scale and Robustness reliability under stress; security mobile agents; PDA devices Agent Frameworks Provide for easy creation of agents Communication Layer Messaging Libraries Libraries for accessing services Directory Mediation Security Remittance Operations Agent Frameworks Example: JATLite (~1998) Some Examples Homebrew Academic JATLite (Stanford) IMPACT (UMD-CP) RETSINA (CMU) D Agents (Dartmouth) Industrial EMAA (LMCO ATL) COUGAAR (DARPA Logistics Programs) Aglets (IBM) 51 Java-based rapid creation of agents Enables Communication and Interaction Knowledge Query and Manipulation Language (KQML) over TCP/IP Other features left to developers Basic Ideas: Agent Message Routers (AMR) Typed-Message Agents 5 JATLite Overall Picture Example: RETSINA Agents register with AMR KQML Messages routed through AMR Agent 1 does not need to know Agent location 53 Reusable Environment for Task-Structured Intelligent Networked Agents From K. Sycara et CMU Basic concepts: Interface agents -- interact with users, receive user input, and display results. Task agents -- help users perform tasks, formulate problem-solving plans and carry out these plans by coordinating and exchanging information with other software agents. Information agents -- provide intelligent access to a heterogeneous collection of information sources. Middle agents -- help match agents that request services with agents that provide services. 54 9

10 The Communication and Coordination module accepts and interprets messages and requests from other agents. The Planning module takes as input a set of goals and produces a plan that satisfies the goals. The Scheduling module uses the task structure created by the planning module to order the tasks. The Execution module monitors this process and ensures that actions are carried out in accordance with computational and other constraints. RETSINA Agents Agents collaborate on the network (or CoABS grid) Agents interact with users Agent perform distributed planning and control RETSINA Enabling Agents: A Word on Grid Computing Computing Power treated as a utility Like electricity Plug in to access the grid s power Popular examples DES cracking SETI@Home Napster, Gnutella 57 Internet as Agent Testbed Massive data Chaotic Dynamic Conflicting Lots of Agent Aps: information gathering information filtering Infrastructure for building agents and agent grids 58 Example of a Grid: SETI@home SETI: Search for Extraterrestrial Intelligence Users download a program Idle/Background processor time crunches radio data Creates massive multi-t-flop virtual computer SETI@home

11 Building Agent Grids CoABS Grid: Framework for Federating Heterogeneous Agent Systems Requires: run-time interoperability Of diverse systems Of diverse agent types Problem solving agents User Interface agents Broker, visualization, and other control agents Translation agents 61 Enables dynamic interoperability of distributed, heterogeneous, objects, services, and multi-agent systems Flexible run-time communications Dynamic registration & discovery of relevant participants Adaptive and Robust System evolves to meet changing requirements without reconfiguring network Failed or unavailable components purged from the Registry Functionally-similar replacements for failed components easily found Foundation for additional common services provided by CoABS researchers Leverages COTS: Java, Jini, RMI, FIPA, XML 6 CoABS Grid High Level View Agent is a special kind of object internal event loop communicates via message exchange LAN Grid Member Grid LAN agent/object registered with grid services Grid Community set of Grid services LAN and registered members Grid Services infrastructure services to facilitate interoperability across heterogeneous multi-agent systems Case Studies Grid is the set of all registered members and the federation of grid services Case Study: Vehicle Paintshop N paint booths M > N paint colors Trucks arrive from assembly line Requires 1 of M paint colors Cost for switching color of booth Some paint is wasted when system flushed Time Each paint booth has a queue Trucks cannot be reordered in queue FULL? FULL Accessible or Inaccessible? Deterministic or Non-deterministic? Episodic or Non-episodic?

12 Accessible (as described) If you consider the state of other parts of the factory that can affect trucks that arrive later, then inaccessible Deterministic or Non-deterministic? Episodic or Non-episodic? 67 Accessible (as described) If you consider the state of other parts of the factory that can affect trucks that arrive later, then inaccessible E.g., machine failures Episodic or Non-episodic? 68 Accessible (as described) If you consider the state of other parts of the factory that can affect trucks that arrive later, then inaccessible E.g., machine failures Non-episodic E.g., an agent s performance can be affected by the color required of trucks arriving in the future 69 Accessible (as described) If you consider the state of other parts of the factory that can affect trucks that arrive later, then inaccessible E.g., machine failures Non-episodic E.g., an agent s performance can be affected by the color required of trucks arriving in the future Dynamic 70 Accessible (as described) If you consider the state of other parts of the factory that can affect trucks that arrive later, then inaccessible E.g., machine failures Non-episodic E.g., an agent s performance can be affected by the color required of trucks arriving in the future Dynamic Discrete 71 Machine 1 Machine Machine N Simple Reactive Agent: Routing Wasps C A B B response thresholds φ φ,... A, B S P( route ST, φt ) = S φ A C B stimulus S B T T + T 7 1

13 Response Threshold Updates While processing or setting up job type j θ w, j = θ w, j δ1 While processing job type other than j θ w, j = θ w, j + δ While idle for all job types j θ θ w, j = w, j δ 3 73 Another Simple Reactive Agent: Scheduling Wasps Machine Currently processing A A B A A Dominance struggle FC P( success FC, FP ) = F + F B Queue C P Force variable F w Job characteristics Arrival time, due date, priority, setup time, processing time, 74 CMU s Federation of Intelligent Robotic Explorers (FIRE) Project Goal: coordinate team of robots exploring an unknown planetary Case Study: Multi-robot Exploration scientist scientist scientist Problem domain: Multi-rover exploration many scientists bandwidth interface heterogeneous many tasks/goals environment blackouts broker autonomous 75 relative priorities 76 OpTrader limited resources... UI UI... UI rover rover rover... rover simulator Accessible or Inaccessible? Deterministic or Non-deterministic? Episodic or Non-episodic? Inaccessible Deterministic or Non-deterministic? Episodic or Non-episodic?

14 Inaccessible Episodic or Non-episodic? Inaccessible Non-episodic Inaccessible Non-episodic Dynamic Inaccessible Non-episodic Dynamic Continuous 81 8 Distributed Three-Layered Architecture UI / OpTrader Market-based Planning Behavioral Layer Real-time sensor/effector feedback loops Efficient distributed servo loops (sensor/effector connections) Tightly coordinated controllers (effector/effector connections) Executive Layer Mediator between planning and behavioral layers Receives distributed plans, translates them, establishes appropriate behavioral connections/functionality plans (PRL) configuration Planning Executive Behavior Rover 1 status / data status / data plans tasks synch inputs outputs Simulator Planning Executive Behavior Rover 83 Planning RoboTrader Scheduler Executive OpTrader Robo Trader Planning Prospective Task Task Feasibility, Cost Awarded Task Execute Task Scheduler T 1 T T 3 Executive Task Status 84 14

15 References Some slides came in whole or in part from the lecture materials associated with the Wooldridge textbook. For more information, see the Wooldridge textbook; The Russell & Norvig AI textbook; a survey article in Volume 1, Issue 1 of the Journal of Autonomous Agents & Multi- Agent Systems