| Collective Intelligence (CI) of Multiagent system (MAS) is the ability of a collection of agents, even with simple intelligence, which can come out with a better solution to a problem, than that of the summation of the abilities of all agents when they work individually. CI is a specific property, the emergent property, of MAS instead of a gathering phenomenon of multiple intelligent entities. Designing MAS with CI properties is different from designing the traditional MAS, in which the problem solving ability or planning are pre-programmed inside the agents. When designing MAS with CI properties, a higher level of problem solving ability, or intelligence can emerge by properly tuning the interaction between agents and adjusting agent behaviors. In this dissertation, the new CI-based MAS design principle is applied to the autonomous collective robotics system and modified particle swarm optimization (PSO) algorithm design.; In the collective robotics applications, we implemented CI principles in three different approaches, a gradual expansion-based exploring approach, a fuzzy logic control approach and a bias expansion swarm based approach, in which multiple simple, low-cost homogenous robots are used to accomplish the task that cannot be fulfilled by any robot when each works individually. We demonstrated that the CI principles could be applied to solve the problem of searching and engaging tasks in large-scale area or finding the emission sources of hazardous materials.; In the domain of PSO research, we proposed the Tracking Dynamical Particle Swarm Optimizer (TDPSO) that can efficiently locate and track the optimal solution in a dynamically changing environment. In TDPSO, the particle's structure is different from traditional PSO. Each particle's knowledge is applied an "evaporation constant" to gradually weaken the knowledge's validity. Through this mechanism, the knowledge of each particle will be gradually updated in a dynamically changing environment. |
