Comportamentos segregativos em enxames robóticos

• Ano de defesa: 2014
• Autor(a) principal: Vinicius Graciano Santos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Enxames; Segregação; Robótica; Computação
• Link de acesso: http://hdl.handle.net/1843/ESBF-9K9H22

Swarm robotics is the study of large multi-agent systems whose robots are relatively simple and have limited capabilities. These systems usually rely on properties such as robustness, flexibility, and scalability to fulfill complex tasks on distinct scenarios. In order to achieve these properties, robotic swarms generally simulate the collective behavior of insects and animals, which display intricate mechanisms shaped by evolution as solutions to many real-world problems. A basic requirement for most robotic swarms is the ability for safe navigation in shared environments. Particularly, a desired behavior is to avoid merging with different teams navigating in opposite directions. This is an example of segregation, a natural phenomenon which is commonly observed in nature. Several biological systems adopt self-sorting mechanisms based on segregative behaviors. Among these, cell segregation is of particular interest since it plays an important role in the formation of tissues, organs, and living organisms. In this work, we study segregation in swarm systems and propose solutions to two particular problems: segregated clustering and segregated navigation. We tackle the former by exploring the Differential Adhesion Hypothesis, which states that cells naturally segregate because of differences in affinity, and introduce a controller that can segregate heterogeneous swarms of robots according to the characteristics of each agent, such that similar robots form homogeneous teams and dissimilar robots are segregated. Regarding the latter problem, we present a distributed mechanism that combines concepts such as hierarchical abstractions, flocking behaviors, and velocity obstacles in order to maintain teams of robots segregated during navigation. We perform simulated and real experiments in order to study the feasibility and effectiveness of our methods. Results show that our approaches allow a swarm of multiple heterogeneous robots to segregate in a coherent and smooth fashion, without any interagent collisions.