Engineering swarm systems: A design pattern for the best-of-n decision problem

The study of large-scale decentralised systems composed of numerous interacting agents that self-organise to perform a common task is receiving growing attention in several application domains. However, real world implementations are limited by a lack of well-established design methodologies that provide performance guarantees. Engineering such systems is a challenging task because of the difficulties to obtain the micro-macro link: a correspondence between the microscopic description of the individual agent behaviour and the macroscopic models that describe the system's dynamics at the global level. In this thesis, we propose an engineering methodology for designing decentralised systems, based on the concept of design patterns. A design pattern provides a general solution to a specific class of problems which are relevant in several application domains. The main component of the solution consists of a multi-level description of the collective process, from macro to micro models, accompanied by rules for converting the model parameters between description levels. In other words, the design pattern provides a formal description of the micro-macro link for a process that tackles a specific class of problems. Additionally, a design pattern provides a set of case studies to illustrate possible implementation alternatives both for simple or particularly challenging scenarios. We present a design pattern for the best-of-n, decentralised decision problem that is derived from a model of nest-site selection in honeybees. We present two case studies to showcase the design pattern usage in (i) a multiagent system interacting through a fully-connected network, and (ii) a swarm of particles moving on a bidimensional plane.Doctorat en Sciences de l'ingénieur et technologie