Using computational stigmergy for swarmbased data sensing and analysis: from samples to intelligent agents

Thanks to nowdays technology advancement, a large amount of data are generated from many different fields ranging from economy, health monitoring, communications, trasportation management, or robotics. While analyzing this kind of data, one of the main problem is to identify relevant or recurrent temporal patterns and recognize anomalous one. For instance: a relevant change in the trend of social and economic indicators is fundamental for policy makers investment decisions; the reduction of the speed of multiple vehicles can identify a traffic congestion; the changing of the movement patterns of a person can identify the behavioral shift in her/his daily activities; when applied to a sleep monitoring it can help to understand the sleep quality and its progressive degeneration due to subject’s disease. While facing the analysis of such data, there are numerous levels of complexity i.e. the high dimensionality, the hard-to-represent dynamics underlying an observed phenomenon, and the definition of a similarity measure aimed at work with different dynamics without explicitly modelinging them, since such model may work only under the assumption formulated by the designer. In order to tackle those issues, we exploit a biologically-inspired computational approach based on stigmergy. In biology, stigmergy is a form of indirect communication and coordination used by social insects. Specifically, each individual releases a pheromone mark in a shared environment while performing a specific action (e.g. carrying a piece of food). At the same time, its behavior is affected by the pheromones perceived in the environment (e.g. following the pheromones trail towards the source of food). In that way, subsequent actions tend to reinforce and build on each other, leading to the spontaneous emergence of coherent, apparently systematic activities. Finally, this indirectly coordinated activity has a defined temporal extension since the pheromones, given their volatility, evaporate over time. This effect is counteracted only if many pheromones are subsequently deposited in proximity with each others (thus they aggregate), resulting in the appearance of a stable pheromone trail in correspondence of this regular depositing activity. In computer science, stigmergy can be employed as a dynamic, agglomerative, computing paradigm able to embody both spatial and temporal domain. Computational stigmergy focuses on the low level processing, where individual samples are augmented with dynamic micro-structure to enable their spatio-temporal aggregation. Such aggregation summarizes micro and macrodynamics in data, allowing the computation of a degree of similarity between different dynamics. Finally, this approach is specialized for each case study, by employing an adaptation mechanism based on a evolutionary algorithm. Here, different applications of computational stigmergy are studied, showing the feasibility and the capability of such approach to be adopted in heterogeneous fields. Moreover, at the final stage of the architecture development, we compare the proposed approach with state-of-art techniques on classification task. Stigmergy is also used as a self-organization mechanism that can be fruitifully exploited in the context of swarm robotics. Swarm robotics systems have the potential to shape the future of many applications, e.g. targeted material delivery, precision farming, and distributed target search. In this contexts, a virtual representation of the pheromone is used to steer the swarm toward the most convenient part of the scenario, e.g. the area with the higher probability to have the presence of target or material to carry. Here, different applications of stigmergy-based swarm coordination are presented, showing the convenience of such approach both with distributed target search via UAVs and distributed material collection via robot.