A collective intelligence model for swarm robotics applications

Swarm intelligence models represent a powerful tool to address complex tasks by multi-agent systems, although they are rarely used in practical applications as decentralized cooperation logic. Modern challenges include the improvement of model reliability with small swarm sizes and enhancing performance with minimal number of free parameters. Available techniques are generally tuned for computational optimization, at the expense of the applicability to real-world scenarios. Merging concepts from meta-heuristic methods and consensus theory we propose a swarm cooperation model which can act both as virtual optimizer and vehicle controller. The model shows a higher or equal success rate with respect to benchmark methods on 22 out of 33 landscapes when dealing with less equal 16 agents and low dimensional problems. Beyond multimodal optimization, a computational proof of concept shows that the method can successfully drive the contaminant localization in a complex marine environment by controlling a group of autonomous underwater vehicles. The Swarm Cooperation Model (SCM), governing the balance between social interactions, cognitive stimuli and stochastic fluctuations leads an agent swarm to accomplish complex tasks, such as the optimization of multimodal functions or the localization of a contaminant source.