Localização cooperativa e descentralizada em enxames robóticos

• Ano de defesa: 2015
• Autor(a) principal: Anderson Grandi Pires
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Localização Cooperativa; Enxame de robôs; Localização cooperativa; Movimentação Coletiva; Enxame de Robôs; Robótica Probabilística; Sistemas Multirrobóticos; Movimentação coletiva; Robótica; Computação; Robôs Movimento
• Link de acesso: http://hdl.handle.net/1843/ESBF-AAVQNM

A common strategy to provide localization in multi-robot systems is to use part of the group as dynamic landmarks, in order to improve the quality of the pose estimation in a process known as Cooperative Localization. This occur, when two or more robots meet and exchange information with the aim to improve their localization estimate. In Swarm Robotics, the usage of this kind of technique is little explored, possibly because the time and space complexity to generate consistent localization estimates. In this work, we propose an approach to cooperatively localize swarms in a decentralized way. We consider the collective movement of the group, once this can improve the possibility of encounters, which effects the quality of the group localization. The collective movement strategy is also proposed in this work, despite the presented localization method can be applied by making use of other motion strategies. We deal with the complexity related to the cooperative localization of a large number of robots by using the Covariance Intersection Algorithm, which was proved in the literature to generate consistent estimates in situations where the interdependence information is not known. Thus, this work proposes a scalable, decentralized and consistent approach to perform cooperative localization in swarms. Simulations were employed to analyse the scalability and some aspects that can influence in the localization quality. Real experiments were also developed as a proof of concept with the aim to show the feasibility of using this approach in real robots. The results show the effectiveness of the proposed approach.