Abstract
Natural swarms have inspired the development of artificial swarming systems capable of solving a wealth of problems across numerous areas, including robotics, optimization, finance, organizational study, and beyond. The intriguing collective behaviors of eusocial animals—insect colonies, flocks of birds, schools of fish—have left researchers wondering “What governs the swarm? Who provides the instructions to direct and control the swarm?” A key answer is that these systems exhibit a unique form of intelligence, a collective intelligence dubbed swarm intelligence. Swarm Intelligence, a relatively new term, was first coined by Beni and Wang in 1993, while they were working on cellular robotic systems. Different groups of researchers like biologists, roboticists, computer scientists, and physicists have since built upon this definition, from the perspective of their own discipline. This has led to multiple non-unified definitions of the concept of swarm intelligence, which can have an adverse effect on the advancement of swarm intelligence and swarming systems. For instance, apparently simple questions like “Does this system possess swarm intelligence?”, “How (swarm) intelligent is this system?”, and “Is system A more (swarm) intelligent than system B?” are absolutely nontrivial. This points to the lack of a form of quantification of swarm intelligence. On the swarming system design front, recent works have also suggested that there is a need for standardization in its design methodology. A benchmark problem would seem to be a reasonable and much needed solution to enable the comparison of swarm intelligence across various systems, such that further improvements to the system can be made. This thesis aims at: (i) revisiting the concept of swarm intelligence, (ii) proposing a quantification of swarm intelligence, (iii) developing a benchmark framework, and (iv) performing a quantitative study of the transferability of swarm intelligence from one problem to another