Swarm intelligence: tools and applications

该提案的目标是开发理论、工具和技术，以促进使用群体智能解决问题的系统的构建。为了更好地理解基于群的系统中的信息处理，我们提出了一种新的信息场观。我们预计，这将基于研究人员自己的博士论文、班迪尼等人的情境细胞代理人的工作以及我们自己在基于相互作用的信息场理解人群动力学方面的工作的扩展。作为所提出的理论公式的实用性的实际例子，我们期望将群算法应用于机器人足球领域的问题，并继续在理解人群方面的工作。我们在前一个问题中的重点将是动态的角色分配，而在后一个问题中，我们了解如何在人群中更有效地交流信息，以便例如更快地离开大楼。我们希望使用进化算法，如遗传编程，以便以类似于Sarmbot项目的方式为基于群体的机器人开发代理行为，该项目使用神经网络来描述机器人的行为。在理解传感器网络时，我们希望开发算法，确保在单个传感器发生故障的情况下进行可靠的信息收集；同样，进化计算的技术有望提供行为。作为对群处理效用的进一步演示，我们期望解决自主计算领域的问题。在这里，我们预计自主管理者--也就是代理--将学会自组织，使下一代计算机系统的管理自动化。

The objectives of the proposal are to develop theories, tools and techniques that facilitate the construction ofsystems which solve problems using Swarm Intelligence. In order to achieve an understanding of information processing in swarm-based systems, we propose that a new information field view be developed. This, we expect, to be based upon the investigator's own Ph.D. thesis, the work of Situated Cellular Agents of Bandini et al and extensions of our own work in understanding crowd dynamics based upon interacting information fields. As practical examples of the utility of the proposed theoretical formulation, we expect to apply swarm algorithms to problems in the area of robotic soccer and continue work in the understanding of crowds. Our focus in the former problem will be dynamic role assignment, while in the latter problem understanding how  information can be more effectively communicated within a crowd in order to effect, for example, more rapidegress from a building. We expect to use evolutionary algorithms, such as Genetic Programming, in order todevelop agent behaviours for swarm-based robots in a manner similar to the SwarmBot project, which usesneural networks for robot behavior. In understanding sensor networks we expect to develop algorithms thatensure robust information gathering in scenarios where individual sensors fail; again, techniques fromEvolutionary Computation are expected to be provide behaviours. As a further demonstration of the utility ofswarm processing, we expect to solve problems in the area of Autonomic Computing. Here, we anticipate thatautonomic managers -- agents by another name -- will learn to self organize to automate the management ofnext-generation computer systems.