Tutorial Outline Introduction to Autonomous Agents and Multi-Agent Systems I Agents N What are they? N Why are they a good idea? Michael Luck I Agent Architectures Dept of Electronics and Computer Science N Deliberative (especially BDI models) University of Southampton, UK. mml@ecs.soton.ac.uk N Hybrid http://www.ecs.soton.ac.uk/~mml N Reactive I Agent Interactions I Agent Resources Remote Agent RAX Experiment (RAX) Comprises I N planner/scheduler to generate plans for general mission goals I Deep Space One N smart executive to execute plans mission to N Mode identification and recovery validate to detect failures technologies Goals not pre-planned so more I flexible I AI software in primary command Tests include simulated failures I of a spacecraft Tests in May 1999 I Agents Agent Definitions I Relatively new field (10 years?) I Smith et al: “persistent software entity dedicated to a specific I Dramatic growth purpose” I Popularity I Selker: “computer programs that I Increasing numbers of applications simulate a human relationship by I Multi-disciplinary doing something that another I Problems: person could do for you “ N Agent backlash? I Riecken: “integrated reasoning N Sound conceptual foundation? processes” CACM, July 1994 1
Why agents? … and more I Increasingly difficult to deal with large-scale I anything that can be viewed as information systems using traditional software: perceiving its environment through N distributed and open, lacking central control and standardised communication; sensors and acting upon that N heterogeneous: compatibility and interfacing environment through effectors problems; - Russell and Norvig N rapid change: new subsystems appear, existing I An autonomous agent … senses ones disappear that environment and acts on it, N rapid growth: huge amount of unstructured information; over time, in pursuit of its own N human involvement: sophisticated interaction and agenda and so as to affect what it cooperation . senses in the future.'' - Franklin and Graesser Agent Types Application Areas I Agent monitoring of web sites I Software agents I Agent filtering of email and I Interface agents newsgroups I Personal assistant agents I Personal information management I Believable agents I Electronic marketplaces I Electronic mail agents N “an agent is a credit card with an I Information agents attitude” - Richard Sharpe I Teaching agents I Negotiation between and within organisations Agent Dimensions Lack of Agreement I Reactivity I Pro-activeness I Does it matter? I Autonomy I Richness aids acceptance I Rationality I Broad range of applicability I Benevolence I Cross-fertilising subfields I Veracity I Lack of precision I Temporal continuity I Abuse of terminology I Adaptability I Mobility I Social ability 2
Weak Notion of Agents Strong notion of agents I Four key qualities: I In addition to the weak notion, also N Autonomous: function without uses mental components such as intervention N belief N Proactive: goal-directed behaviour N desire N Reactive: perceive and respond to changing environment N intention N Social ability: interaction with others N knowledge N etc - Wooldridge and Jennings, 1994/1995 Objects Agents Autonomous Agents Agent “encapsulated computer system, situated in some environment, and capable of flexible flexible autonomous action in that environment in order to meet its design objectives” (Wooldridge) I control over internal state and over own behaviour I experiences environment through sensors and acts through effectors I reactive: respond in timely fashion to environmental change I proactive: act in anticipation of future goals 18 3
Agent Interactions Multiple Agents I Interaction between agents is inevitable N to achieve individual objectives, to manage inter- In most cases, single agent is insufficient dependencies N no such thing as a single agent system (!?) I Conceptualised as taking place at knowledge-level N which goals, at what time, by whom, what for N multiple agents are the norm, to represent: ! natural decentralisation I Flexible run-time initiation and responses ! multiple loci of control N cf. design-time, hard-wired nature of extant approaches ! multiple perspectives paradigm shift from previous perceptions of ! competing interests computational interaction 19 20 Organisations Organisations Agents act/interact to achieve objectives: This organisational context: N on behalf of individuals/companies N influences agents’ behaviour N part of a wider problem solving initiative ! relationships need to be made explicit • peers underlying organisational relationship • teams, coalitions between the agents • authority relationships N is subject to ongoing change ! provide computational apparatus for creating, maintaining and disbanding structures 21 22 A Canonical View Decomposition: Agents Agent Organisational In terms of entities that have: I Interactions relationships N own persistent thread of control (active: “say go”) N control over their own destiny (autonomous: “say no”) Makes engineering of complex systems easier: I N natural representation of multiple loci of control ! “real systems have no top” N allows competing objectives to be represented and Environment reconciled in context sensitive fashion Sphere of influence 4
Decomposition: Interactions Complex System Agent-Based System I Agents make decisions about nature & scope of Sub-systems interactions at run time I Makes engineering of complex systems easier: Sub-system components N unexpected interaction is expected ! not all interactions need be set at design time Interactions between sub-systems and sub-system components N simplified management of control relationships between components Relationships between sub-systems ! coordination occurs on as-needed basis between and sub-system components continuously active entities Complex System Agent-Based System Complex System Agent-Based System Sub-systems Agent organisations Sub-systems Agent organisations Sub-system components Sub-system components Agents Interactions between sub-systems and Interactions between sub-systems and sub-system components sub-system components Relationships between sub-systems Relationships between sub-systems and sub-system components and sub-system components Agents Consistent with Complex System Agent-Based System Trends in Software Sub-systems Agent organisations Engineering Conceptual basis rooted in problem domain I Sub-system components Agents Increasing localisation and encapsulation I N apply to control, as well as state and behaviour “cooperating to achieve common objectives” Interactions between sub-systems and sub-system components “coordinating their actions” “negotiating to resolve conflicts” Relationships between sub-systems and sub-system components Explicit mechanisms for representing & - change over time managing organisational relationships Structures for modelling collectives - treat collections as single coherent unit 5
Agents Consistent with Agents Support System Trends in Software Development by Engineering Synthesis Conceptual basis rooted in problem domain I An agent is a stable intermediate form N able to operate to achieve its objectives and interact with others Increasing localisation and encapsulation I in flexible ways construct “system” by bringing agents together and watching Greater support for re-use of designs and programs I overall functionality emerge from their interplay N whole sub-system components (cf. components, patterns) ! e.g. agent architectures, system structures N well suited to developments in: N flexible interactions (cf. patterns, architectures) ! open systems (e.g. Internet) ! e.g. contract net protocol, auction protocols ! e-commerce Single Agent Single-Agent Architectures Architectures I Deliberative Agent Systems N Symbolic representation and manipulation BDI N IRMA, GRATE, PRS/ dMARS PRS/dMARS I Reactive N Stimulus -Response Agent Systems N Subsumption Architecture N Agent Network Architecture I Hybrid Agent Systems N Act both deliberatively and reactively N TouringMachine N InterRRaP Towards BDI BDI Systems Architectures I BDI aims to model rational or intentional I BDI = Belief, Desires and agency Intentions I The symbols representing the world I Many agent architectures are BDI correspond to mental attitudes based I Three categories: I Original system was PRS N informative (knowledge, belief, I More recent versions include assumptions) dMARS. N motivational (desires, motivations, goals) I Other related systems include N deliberative (intentions, plans) AgentSpeak(L) and Agentis 6
Recommend
More recommend
