In this research, the application of intelligent mobile agents to the management of distributed network environments is investigated. Intelligent mobile agents are programs which can move about network systems in a deterministic manner in carrying their execution state. These agents can be considered an application of distributed artificial intelligence where the (usually small) agent code is moved to the data and executed locally. The mobile agent paradigm offers potential advantages over many conventional mechanisms which move (often large) data to the code, thereby wasting available network bandwidth. The performance of agents in network routing and knowledge acquisition has been investigated and simulated. A working mobile agent system has also been designed and implemented in JDK 1.2.

published_or_final_version / abstract / Industrial and Manufacturing Systems Engineering / Doctoral / Doctor of Philosophy

published_or_final_version / Industrial and Manufacturing Systems Engineering / Master / Master of Philosophy

published_or_final_version / Industrial and Manufacturing Systems Engineering / Master / Master of Philosophy

by Zhao Lei. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2002. / Includes bibliographical references (leaves 78-85). / Abstracts in English and Chinese. / Abstract --- p.ii / Acknowledgments --- p.iv / Chapter 1 --- Introduction --- p.1 / Chapter 2 --- Background --- p.5 / Chapter 2.1 --- Agents --- p.5 / Chapter 2.1.1 --- Deliberative Agents --- p.7 / Chapter 2.1.2 --- Reactive Agents --- p.8 / Chapter 2.1.3 --- Interacting Agents --- p.9 / Chapter 2.1.4 --- Hybrid Architectures --- p.10 / Chapter 2.2 --- Real-time Artificial Intelligence --- p.10 / Chapter 2.3 --- Real-Time Agents --- p.12 / Chapter 2.3.1 --- The Subsumption Architecture --- p.13 / Chapter 2.3.2 --- The InterRAP Architecture --- p.15 / Chapter 2.3.3 --- The 3T Architecture --- p.16 / Chapter 2.4 --- On-line Scheduling in Real-Time Agents --- p.18 / Chapter 3 --- A Real-Time Agent Architecture --- p.20 / Chapter 3.1 --- Human Cognition Model --- p.20 / Chapter 3.1.1 --- Perception --- p.22 / Chapter 3.1.2 --- Cognition --- p.22 / Chapter 3.1.3 --- Action --- p.23 / Chapter 3.2 --- Real-Time Message Passing Primitives and Process Structuring --- p.24 / Chapter 3.2.1 --- Message Passing as IPC --- p.25 / Chapter 3.2.2 --- Administrator and Worker Processes --- p.28 / Chapter 3.3 --- Agent Architecture --- p.29 / Chapter 3.3.1 --- Sensor Workers and the Sensor Administrator --- p.30 / Chapter 3.3.2 --- The Cognition Workers --- p.32 / Chapter 3.3.3 --- "The Task Administrator, the Scheduler Worker and Ex- ecutor Workers" --- p.32 / Chapter 3.4 --- An Agent-Based Real-time Arcade Game --- p.34 / Chapter 4 --- A Multiple Method Approach to Task Scheduling --- p.37 / Chapter 4.1 --- Task Scheduling Mechanism --- p.37 / Chapter 4.1.1 --- Task and Action --- p.38 / Chapter 4.1.2 --- Task Administrator --- p.40 / Chapter 4.1.3 --- Task Scheduler --- p.43 / Chapter 4.2 --- A Task Scheduling Model --- p.44 / Chapter 4.3 --- Combination Rules and Special Cases --- p.46 / Chapter 4.4 --- Scheduling Algorithms --- p.49 / Chapter 5 --- Task Scheduling Model: Analysis and Experiments --- p.53 / Chapter 5.1 --- Goodness Measure --- p.53 / Chapter 5.2 --- Theoretical Analysis --- p.54 / Chapter 5.3 --- Implementation --- p.59 / Chapter 5.3.1 --- Task Generator Implementation --- p.59 / Chapter 5.3.2 --- Executor Workers Implementation --- p.61 / Chapter 5.4 --- Experimental Results --- p.62 / Chapter 5.4.1 --- Hybrid Mechanism and Individual Algorithms --- p.63 / Chapter 5.4.2 --- Effect of Average Execution Time --- p.65 / Chapter 5.4.3 --- Effect of the Greedy Algorithm --- p.65 / Chapter 5.4.4 --- Effect of the Advanced Algorithm --- p.67 / Chapter 5.4.5 --- Effect of Actions and Relations Among Them --- p.68 / Chapter 5.4.6 --- Effect of Deadline --- p.71 / Chapter 6 --- Conclusions --- p.73 / Chapter 6.1 --- Summary of Contributions --- p.73 / Chapter 6.2 --- Future Work --- p.75

Intelligent robots are rational agents. The rationality of robots working cooperatively is significantly different from robots working independently. Cooperation between intelligent robots requires the high level of reasoning and complex interactions for successful operations. The required reasoning process includes knowledge representation and sharing as well as the ability to understand the context of a situation. The reasoning process heavily influences on the planning of deciding what actions need to be taken. Goal deliberation and planning is the process that deals with those requirements. This dissertation investigates the problem of goal deliberation and planning to enable such cooperation between goal-oriented intelligent robots, working as a team. The dissertation then proposes a multi-robot system model that embraces results of the investigation. The proposed model is realized on the top of the platform ‘robot operating system’ (ROS). The implemented system, named ‘goal-oriented multi agent systems’ (GOMAS), is demonstrated with the computer game, StarCraft II. Units in StarCraft II are individually controlled by the GOMAS robots and work cooperatively to attain a set of goals given from operators. The demonstration with the three different scenarios validates that the GOMAS system successfully and efficiently deliberates and plans the given goals.

Esta dissertação apresenta uma arquitetura de controle distribuída para um sistema de visão computacional propositada, modelada segundo uma abordagem de Inteligência Artificial Distribuída com ênfase em Sistemas Multi-Agentes. O principal problema abordado é o da integração de diversos módulos com tarefas e comportamentos diferentes em um sistema robótico, que utiliza visão computacional para perceber o mundo. Esta integração surge da organização de diversas soluções dedicadas a diferentes tarefas visuais, resultando na proposta de uma arquitetura constituída por uma sociedade de Agentes Autônomos (AAs), que comunicam entre si através de uma rede de comunicação descentralizada e totalmente conectada, sendo cada AA responsável por um comportamento independente, organizados segundo regras de comportamento e uma estrutura de autoridade. Essa arquitetura integra diversos módulos da cognição, como percepção visual, planejamento, controle e atuação. Como tentativa de validação experimental da proposta, essa arquitetura foi implementada em um sistema dotado de um manipulador robótico e uma câmera, trabalhando em um domínio de montagem. A estrutura básica da arquitetura foi desenvolvida segundo uma metodologia de projeto orientada a objetos e consiste em uma biblioteca de classes que definem o modelo dos agentes e as estruturas para as interações entre esses em uma sociedade. O sistema robótico proposto foi implementado na Célula Flexível de Montagem da Escola Politécnica da USP, tendo por objetivo realizar tarefas de montagens simples e visualmente guiadas. O sistema foi testado exaustivamente e os resultados experimentais indicaram algumas vantagens e desvantagens da metodologia empregada, possibilitando a definição de algumas diretrizes importantes para o projeto e implementação de sistemas robóticos inteligentes atuando no mundo real. / This dissertation presents a distributed control architecture for a purposive computer vision system, modeled according to a Distributed Artificial Intelligence approach, with emphasis in Multi-Agents Systems. The main problem addressed is the integration of several modules with different purposes and behaviors in a robotic system, which uses computer vision to perceive the world. This integration arises from the organization of several dedicated solutions to different visual tasks, resulting in the proposal of one architecture that consists of a society of Autonomous Agents, communicating with each other through a decentralized and fully connected network, where each AA is responsible for an independent behavior. This society is organized according to behavior rules and an authority structure. Furthermore, this architecture integrates several cognitive modules, like visual perception, planning, control and action. As an attempt of an experimental validation of the proposal, this architecture was implemented in a system composed of a robotic manipulator and one camera, working in an assembly domain. The basic structure of the architecture was developed according to an object oriented methodology and consists of a library of classes which defines the agents\' model and the structures needed for the interactions among these agents in a society. The proposed robotic system was implemented in the Flexible Assembly Cell of the Escola Politécnica da USP, with the objective of accomplishing simple and visually guided assembly tasks. The system was thoroughly tested and the experimental results indicated some advantages and inconveniences of the methodology, allowing the definition of some important guidelines for the design and implementation of intelligent robotic systems acting in the real world.

Esta dissertação apresenta uma arquitetura de controle distribuída para um sistema de visão computacional propositada, modelada segundo uma abordagem de Inteligência Artificial Distribuída com ênfase em Sistemas Multi-Agentes. O principal problema abordado é o da integração de diversos módulos com tarefas e comportamentos diferentes em um sistema robótico, que utiliza visão computacional para perceber o mundo. Esta integração surge da organização de diversas soluções dedicadas a diferentes tarefas visuais, resultando na proposta de uma arquitetura constituída por uma sociedade de Agentes Autônomos (AAs), que comunicam entre si através de uma rede de comunicação descentralizada e totalmente conectada, sendo cada AA responsável por um comportamento independente, organizados segundo regras de comportamento e uma estrutura de autoridade. Essa arquitetura integra diversos módulos da cognição, como percepção visual, planejamento, controle e atuação. Como tentativa de validação experimental da proposta, essa arquitetura foi implementada em um sistema dotado de um manipulador robótico e uma câmera, trabalhando em um domínio de montagem. A estrutura básica da arquitetura foi desenvolvida segundo uma metodologia de projeto orientada a objetos e consiste em uma biblioteca de classes que definem o modelo dos agentes e as estruturas para as interações entre esses em uma sociedade. O sistema robótico proposto foi implementado na Célula Flexível de Montagem da Escola Politécnica da USP, tendo por objetivo realizar tarefas de montagens simples e visualmente guiadas. O sistema foi testado exaustivamente e os resultados experimentais indicaram algumas vantagens e desvantagens da metodologia empregada, possibilitando a definição de algumas diretrizes importantes para o projeto e implementação de sistemas robóticos inteligentes atuando no mundo real. / This dissertation presents a distributed control architecture for a purposive computer vision system, modeled according to a Distributed Artificial Intelligence approach, with emphasis in Multi-Agents Systems. The main problem addressed is the integration of several modules with different purposes and behaviors in a robotic system, which uses computer vision to perceive the world. This integration arises from the organization of several dedicated solutions to different visual tasks, resulting in the proposal of one architecture that consists of a society of Autonomous Agents, communicating with each other through a decentralized and fully connected network, where each AA is responsible for an independent behavior. This society is organized according to behavior rules and an authority structure. Furthermore, this architecture integrates several cognitive modules, like visual perception, planning, control and action. As an attempt of an experimental validation of the proposal, this architecture was implemented in a system composed of a robotic manipulator and one camera, working in an assembly domain. The basic structure of the architecture was developed according to an object oriented methodology and consists of a library of classes which defines the agents\' model and the structures needed for the interactions among these agents in a society. The proposed robotic system was implemented in the Flexible Assembly Cell of the Escola Politécnica da USP, with the objective of accomplishing simple and visually guided assembly tasks. The system was thoroughly tested and the experimental results indicated some advantages and inconveniences of the methodology, allowing the definition of some important guidelines for the design and implementation of intelligent robotic systems acting in the real world.

Most artificial neural networks suffer from the problem of catastrophic forgetting, where previously learnt information is suddenly and completely lost when new information is learnt. Memory in real neural systems does not appear to suffer from this unusual behaviour. In this thesis we discuss the problem of catastrophic forgetting in Hopfield networks, and investigate various potential solutions. We extend the pseudorehearsal solution of Robins (1995) enabling it to work in this attractor network, and compare the results with the unlearning procedure proposed by Crick and Mitchison (1983). We then explore a familiarity measure based on the energy profile of the learnt patterns. By using the ratio of high energy to low energy parts of the network we can robustly distinguish the learnt patterns from the large number of spurious "fantasy" patterns that are common in these networks. This energy ratio measure is then used to improve the pseudorehearsal solution so that it can store 0.3N patterns in the Hopfield network, significantly more than previous proposed solutions to catastrophic forgetting. Finally, we explore links between the mechanisms investigated in this thesis and the consolidation of newly learnt material during sleep.

The goal of this thesis is twofold: first, we want to present the distributed belief accommodation and revision model for multi-agent systems that has been developed and, second, we wish to show its applicability to an appropriate domain. The Distributed Belief Accommodation & Revision model, called DeBAteR model, was developed for co-operative heterogeneous multi-agent systems used to model inherently dynamic distributed problems. In these systems, although the agents are able to detect changes both in the environment and in the problem specifications, each agent has only a partial view of the global picture. As a result the information that represents the current state of affairs is dynamic, incomplete and sometimes uncertain. This non-monotonic kind of data is called beliefs ? a belief is a piece of data that is held as correct as long as no contradicting evidence is found or presented. Each agent is expected to include an assumption based truth maintenance module for representing properly this type of data. Our main effort was concentrated on the task of maintaining the system's information, which consists of updating, revising and accommodating the represented beliefs. Belief updating is necessary for including the changes detected by the agents both in the environment and/or in the problem specifications. Belief revision is essential for solving the inconsistencies detected among the represented beliefs. Belief accommodation and revision is crucial for integrating the multiple disparate perspectives regarding the same data items, which may occur whenever there is overlap of expertise domains between the agents. In order to solve the information conflicts that result from the detection of inconsistencies between distinct beliefs or within multi-perspective beliefs we conceived the DeBATeR model. The DeBAteR is fully distributed, provides individual belief autonomy and is made of two methodologies: the pro-active belief accommodation and revision methodology and the delayed belief revision methodology. Whilst the first methodology is used to solve domain independent conflicts, the second methodology was devised for solving domain dependent conflicts. Both methodologies use argumentation for, in the case of the domain independent conflicts, choosing the most credible perspective between the existing multiple perspectives of a belief, and, in the case of the domain dependent conflicts, finding the best alternative belief support set for the affected concepts. These methodologies are distributed and their scope may be internal or collective. The DeBAteR model main contributions are: (i) the pro-active methodology conceived for solving domain independent conflicts and (ii) the capability, not only to represent and maintain individual beliefs and joint beliefs, but also to accommodate, rationally maintain and make use of multi-perspective beliefs. Finally, we describe the developed decision support multi-agent system for choosing adequate project locations, called DIPLOMAT ? Dynamic and Interactive Project Location Test bed, which has the ability of accommodating and revising the represented beliefs according to the DeBAteR model methodologies. / O objectivo desta dissertação é duplo: por um lado, pretendemos dar a conhecer o modelo de revisão e acomodação de crenças para sistemas multi-agente por nós desenvolvido e, por outro, procuramos ilustrar a sua validade descrevendo a aplicação que realizámos. O modelo, designado DeBAteR ? Distributed Belief Accommodation & Revision, destina-se a sistemas multi-agente cooperativos e heterogéneos que modelam problemas inerentemente distribuídos e dinâmicos. Neste tipo de sistemas, a informação que representa o ambiente é dinâmica (os agentes possuem a capacidade de constatar alterações no ambiente e/ou nas condições do problema) e, muitas vezes, incompleta (os agentes possuem visões parcelares da realidade) e/ou incerta. Este tipo de informação, de carácter não definitivo, designa-se por crenças ? uma crença é uma convicção tida como correcta enquanto não for posta em causa por alguma evidência. Cada agente, a fim de representar e manipular crenças, foi enriquecido com um módulo específico de manutenção de consistência baseado em suposições. O nosso esforço concentrou-se na tarefa de manutenção (actualização, revisão e acomodação) da informação do sistema. A actualização de crenças é essencial para incorporar as alterações que os agentes detectam no ambiente e/ou nas condições do problema. A revisão de crenças é indispensável para resolver inconsistências (conflitos) entre as crenças representadas. A acomodação e revisão simultânea de crenças é imprescindível para a integração das múltiplas perspectivas díspares que surgem em relação a um mesmo item de informação (crenças pluri-perspectiva) quando existe sobreposição de domínios de especialidade entre os agentes. Para tentar solucionar estes conflitos entre crenças concebemos um modelo distribuído que assegura autonomia individual de crença. O modelo de acomodação e revisão de crenças DeBAteR é composto por duas metodologias: a metodologia pró-activa de acomodação e revisão de crenças pluri-perspectiva e a metodologia retardada de revisão de crenças. A primeira, destina-se a resolver conflitos independentes do domínio e a segunda destina-se à resolução de conflitos dependentes do domínio. Estas metodologias são suportadas quase integralmente por um sistema de argumentação que procura, no caso dos conflitos independentes do domínio, escolher a perspectiva mais credível e, no caso dos conflitos dependentes do domínio, encontrar o melhor conjunto alternativo de suporte para os conceitos afectados. Esta actividade é descentralizada e pode decorrer quer no âmbito intra-agente, quer no âmbito inter-agente. É ainda de realçar: (i) o carácter pró-activo da resolução dos conflitos independentes do domínio (crenças pluri-perspectiva) e (ii) a capacidade de, não só, representar e manter crenças de âmbito individual (crenças suportadas por apenas um agente) e crenças conjuntas (crenças suportadas por vários agentes), mas também, de sintetizar, manter racionalmente e utilizar crenças pluri-perspectiva. Por último, descrevemos o sistema multi-agente de apoio à decisão no domínio da localização de empreendimentos desenvolvido, denominado DIPLOMAT ? Dynamic and Interactive Project Location Test bed, o qual possui a capacidade de acomodar e rever crenças de acordo com as metodologias concebidas no âmbito do modelo DeBAteR.