Exploring communication network structures for the adaptability of a team of robots working in dynamical foraging environments

• Ano de defesa: 2018
• Autor(a) principal: Viedman, Juan Manuel Nogales
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Ciência da Computação
• Palavras-chave em Português: Foraging; Multi-robot systems; Probabilistic decision-making; Adaptive learning; Task allocation; Task partitioning; Computação; Robótica; Robôs-movimento; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::MATEMATICA DA COMPUTACAO::MODELOS ANALITICOS E DE SIMULACAO
• Link de acesso: https://repositorio.ufu.br/handle/123456789/22758; http://dx.doi.org/10.14393/ufu.te.2018.816

In nature, foraging is a fundamental task for several organisms that exhibit collective behaviors. For this reason, one of the most researched tasks in swarm robotics is foraging. In this work, we consider a team of foraging robots where each member repeats the same task: to pick up an object and transport it to a nest. Robot navigation depends completely on local visual clues. Neither dead reckoning nor global orientation helped them. The controller is a hierarchical state machine, which underlies robot behaviors. They learn which behavior fits better with the current condition. Since the environment changes, we developed strategies that enabled robots to achieve a good adaptation. These strategies dealt with distributed environments and exploited: robot-environment and robot-robot connections. In particular, such set of connections allowed the information to flow at a good rate and, as a consequence, their social learning. Since information influences robot behaviors and team performance, the proposed strategies were evaluated in two scenarios: $i$) task allocation and $ii$) task partitioning. In the task-allocation environment, robots interact within an environment, which consists of three regions where robots can forage. Each region includes a set of interfaces generating objects at different rates. These interfaces share information among them and with the subgroup of robots working inside each region. The proposed strategy reached a near-optimal solution because robots incorporated that information to make their decisions. In the task-partitioning environment, robots employ a set of interfaces to transfer objects between both regions or they carry the objects through an alternative path. In this scenario, those interfaces change their speed for transferring objects and robots only share information with other robots. The proposed solution allowed robots to adjust their connections to adapt their learning rate. As a consequence, they make decisions that fitted better to the changes in the interfaces.