Final Report Summary - E-SWARM (Engineering Swarm Intelligence Systems)
The E-SWARM project has generated results that allow a better understanding of the functioning of swarm intelligence systems and has contributed concretely towards the definition of an engineering methodology for the design and implementation of artificial swarm systems.
The swarm intelligence research discipline studies natural and artificial systems composed of many individuals that coordinate using decentralised control and self-organisation. In particular, the discipline focuses on the collective behaviours that result from the local interactions of the individuals with each other and with their environment.
Swarm intelligence research has a marked multidisciplinary character since systems with the above mentioned characteristics can be observed in a variety of domains. Examples of systems studied by swarm intelligence are colonies of ants and termites, schools of fish, flocks of birds, herds of land animals. Some human artefacts also fall into the domain of swarm intelligence, notably some multi-robot systems, and also certain computer programs that are written to tackle optimisation and data analysis problems. The E-SWARM project is devoted to the study of such artificial systems.
Another way of looking at swarm intelligence research is based on the goals that are pursued, that can be of a scientific or an engineering nature. We had therefore two research threads, both addressed in the E-SWARM project: a scientific thread, whose goal is to model swarm intelligence systems and to single out and understand the mechanisms that allow a swarm intelligence system as a whole to behave in a coordinated way as a result of local individual-individual and individual-environment interactions; and an engineering thread, whose goal was to exploit the understanding developed by the scientific thread in order to design systems that are able to solve problems of practical relevance.
The scientific questions that we have addressed in E-SWARM, and for which we have provided a number of answers, are the following: how does a swarm make collective decisions? What are the costs and benefits of behavioural specialisation in a swarm? How to measure and estimate the performance of a swarm intelligence system? To answer these questions we have developed mathematical models of swarm intelligence systems and have studied their theoretical properties, or analysed the results of simulation experiments.
In E-SWARM we have also been concerned with a number of engineering questions. For example, how to design a swarm intelligence system so that its behaviour satisfies a set of required properties? How to generate the behaviour of the individuals in a swarm in an automatic way? How to let a swarm allocate itself to different tasks in an autonomous way? How to let swarms interact with some kind of supervisor, be it a robot with different capabilities or a human being? How to let a swarm of physical agents cooperate to solve a problem that each agent could not solve alone? All these questions have been addressed by means of case studies where swarm of individuals have been given particular tasks to be solved.
The results of all these studies provide answers to scientific and engineering questions concerning both how artificial swarms function and how they should be designed. We have produced a number of mathematical models of artificial swarm systems and we have used these models to study and measure the performance of many different types of swarm behaviours such as collective decision making, task allocation and task partitioning, flocking, and self-assembly. We have proposed a methodology called property-driven design that supports the design and implementation of swarms with given properties. We have proposed ways of modelling a swarm that facilitate human-swarm interaction. And we have proposed AutoMode, a novel methodology for the automatic design and implementation of swarm controllers. Finally, we have proposed methods to automatically tune the parameters of swarm intelligence algorithms and we have shown that such methods greatly increase the performance of swarm intelligence algorithms.
Our results are building blocks towards the ultimate goal of our still ongoing research: the development of an engineering methodology for the design and implementation of artificial swarm intelligence systems. With E-SWARM we have shown that it is possible to make of swarm intelligence an engineering discipline. In other words, we have shown that it is possible to substitute the design and implementation approach based on “researcher experience” and on experimental trial and error with rigorous methodologies that make use of mathematical and analytical tools.
The swarm intelligence research discipline studies natural and artificial systems composed of many individuals that coordinate using decentralised control and self-organisation. In particular, the discipline focuses on the collective behaviours that result from the local interactions of the individuals with each other and with their environment.
Swarm intelligence research has a marked multidisciplinary character since systems with the above mentioned characteristics can be observed in a variety of domains. Examples of systems studied by swarm intelligence are colonies of ants and termites, schools of fish, flocks of birds, herds of land animals. Some human artefacts also fall into the domain of swarm intelligence, notably some multi-robot systems, and also certain computer programs that are written to tackle optimisation and data analysis problems. The E-SWARM project is devoted to the study of such artificial systems.
Another way of looking at swarm intelligence research is based on the goals that are pursued, that can be of a scientific or an engineering nature. We had therefore two research threads, both addressed in the E-SWARM project: a scientific thread, whose goal is to model swarm intelligence systems and to single out and understand the mechanisms that allow a swarm intelligence system as a whole to behave in a coordinated way as a result of local individual-individual and individual-environment interactions; and an engineering thread, whose goal was to exploit the understanding developed by the scientific thread in order to design systems that are able to solve problems of practical relevance.
The scientific questions that we have addressed in E-SWARM, and for which we have provided a number of answers, are the following: how does a swarm make collective decisions? What are the costs and benefits of behavioural specialisation in a swarm? How to measure and estimate the performance of a swarm intelligence system? To answer these questions we have developed mathematical models of swarm intelligence systems and have studied their theoretical properties, or analysed the results of simulation experiments.
In E-SWARM we have also been concerned with a number of engineering questions. For example, how to design a swarm intelligence system so that its behaviour satisfies a set of required properties? How to generate the behaviour of the individuals in a swarm in an automatic way? How to let a swarm allocate itself to different tasks in an autonomous way? How to let swarms interact with some kind of supervisor, be it a robot with different capabilities or a human being? How to let a swarm of physical agents cooperate to solve a problem that each agent could not solve alone? All these questions have been addressed by means of case studies where swarm of individuals have been given particular tasks to be solved.
The results of all these studies provide answers to scientific and engineering questions concerning both how artificial swarms function and how they should be designed. We have produced a number of mathematical models of artificial swarm systems and we have used these models to study and measure the performance of many different types of swarm behaviours such as collective decision making, task allocation and task partitioning, flocking, and self-assembly. We have proposed a methodology called property-driven design that supports the design and implementation of swarms with given properties. We have proposed ways of modelling a swarm that facilitate human-swarm interaction. And we have proposed AutoMode, a novel methodology for the automatic design and implementation of swarm controllers. Finally, we have proposed methods to automatically tune the parameters of swarm intelligence algorithms and we have shown that such methods greatly increase the performance of swarm intelligence algorithms.
Our results are building blocks towards the ultimate goal of our still ongoing research: the development of an engineering methodology for the design and implementation of artificial swarm intelligence systems. With E-SWARM we have shown that it is possible to make of swarm intelligence an engineering discipline. In other words, we have shown that it is possible to substitute the design and implementation approach based on “researcher experience” and on experimental trial and error with rigorous methodologies that make use of mathematical and analytical tools.