Structure and robustness of multi-agent systems operating in open world environments

目的开发设计方法和功能，配置分布式、自治、多智能体系统的结构元素，以增加其对环境和智能体可变性的鲁棒性。该项目考虑了在现实世界中具有自主操作的复杂系统的鲁棒性。例如系统可以是自主运输网络、计算机数据中心或物联网。它旨在为这样的系统提供一种设计方法，使不必要的紧急行为最小化。预期的方法是将多代理建模和分析应用于复杂系统，特别强调包括自主和智能功能的系统。后者创造了扩展预期方法的潜力，以包括人工进化等学习技术。活动被划分为两个工作主体。首先是研究同质与异质系统元素对不良行为产生的影响。第二部分的工作集中在补偿功能的设计上。为了研究异构系统元素的影响，将设计和模拟一系列与现实世界问题和现有文献相关的复杂系统场景。这些场景将被设计成鼓励紧急行为。例如，在放牧试验中，牧羊人、一群狗和一群羊相互作用，或者在排序练习中，机器人团队承担角色，接受指令，并通过遗传系统共享信息。在这个项目的上下文中，结构元素是多智能体系统的构建块。例如，通信、代理的角色或管理代理之间交互的规范。这些结构元素对于所有代理都是相同的（同质），或者在代理之间是不同的（异质）。本项目假设存在这些元素的组合，它们实现了相同的目标，但不太容易受到来自现实世界条件的不必要的系统行为的影响。它提出了这样的问题：结构要素、异质性和对环境和主体可变性的稳健性之间的联系是什么？如何选择这些结构要素，使MAS对大范围的可变性具有鲁棒性？如何修改代理级的行为，以便在运行时重新配置结构元素，以保持积极的系统级功能？目标1。使用有界接口方法指定多智能体系统中智能体之间的交互，以推断哪些结构元素有可能影响不想要的行为。2. 量化多主体结构（例如，角色、层次结构或规范）的能力，以减少由可变性引起的系统级中断。这将使用基于计算机的多智能体仿真进行研究。在代理层创建动态的、智能的行为，重新配置系统结构，以响应系统层的负面紧急行为。动机工程自主多智能体系统正变得越来越复杂，并依赖于其基本功能。例如运输/社会/金融网络以及在仓库中操作或探索未知环境的人-机器人团队。通常在这些多智能体示例中，系统随着时间的推移随着多个设计师和工程师的输入而发展。没有一个单一的设计权威，在设计阶段不完全知道运行条件。现有的工作通常集中在单个代理的正确功能上。缺乏考虑确保整个系统按预期工作的问题的工作。特别是，可变性的影响和如何设计结构元素来应对未知的条件。

AimTo develop design approaches and functionality which configure the structural elements of distributed, autonomous, multi-agent systems to increase their robustness to environmental and agent variability. High level descriptionThis project considers robustness of complex systems with autonmy operating in the real world. Example systems could be autonomous transport networks, computer data centres or the Internet of Things. It aims to contribute to a design methodoly for such systems which minimises unwanted emergent behaviours. The intended approach is to apply multi-agent modelling and analysis to complex systems, with a particular emphasis on systems which include autonomous and intelligent functionality. The latter creates the potential to extend the intended approach to include learning techniques such as artifical evolution. Activity is intended to be split between two bodies of work. The first is investigating the effect of homogenous vs heterogenious system elements on the emergence of unwanted behaviour. The second body of work focuses on the design of a compensating function. To investigate the effect of heterogenious system elements, a series of complex system scenarios, linked to real world problems and existing literature, will be designed and simulated. These scenarios will be designed to encourage emergent behaviour. For example, sheep herding in which a shepherd, team of dogs and flock of sheep interact during a herding trial, or a sorting exercise in which teams of robots undertake roles, receive instructions and share information through a heirachy.Research QuestionsIn the context of this project, a structural element is the building block of a multi-agent system. For example, communications, the role of an agent or the norms which govern the interaction between agents. These structual elements can be either the same for all agents (homogeneous) or vary between agents (heterogenious). This project hyptothosises that there are combinations of these elements which achieve the same goal but are less susceptable to unwanted system behaviour occuring from real world conditions. It asks the questions: What is the link between the stuctual elements, heterogeniality and the robustness to environmental and agent variability? How can these structual elements be chosen such that the MAS is robust to a wide range of variability? How can behaviour at the agent level be modified such that the structual elements are reconfigured during run time to maintain positive, system level, functionality? Objectives 1. Specify the interactions between agents in a multi-agent system using a bounded interface approach to infer where structural elements have the potential to influence unwanted behaviour. 2. Quantify the capability of multi-agent structures (for example, roles, hierarchies or norms) to reduce system level disruption caused by variability. This will be investigated using computer based multi-agent simulations.3. Create dynamic, intelligent, behaviour at the agent level which reconfigures the system structure in response to negative emergent behaviour at the system level.MotivationEngineered autonomous multi-agent systems are becoming increasingly complex and relied upon for essential functionality. For example transport/social/financial networks and human-robot teams operating in warehouses or exploring unknown environments. Often in these multi-agent examples, the system evolves over time with inputs by multiple designers and engineers. There is not a single design authority and the operating conditions are not completely known at the design stage.Existing work often focuses on the correct function of a single agent. There is a lack of work which considers the problem of ensuring the system as a whole works as intended. In particular, the effect of variability and how the structal elements can be designed to cope with unknown conditions.