Um arcabouço para planejamento e controle descentralizado de robôs heterogêneos

• Ano de defesa: 2012
• Autor(a) principal: Armando Alves Neto
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Cooperação aéreo-terrestre; Planejamento multirrobôs; Controle descentralizado; Árvores aleatórias de exploração rápida; Robótica; Computação
• Link de acesso: http://hdl.handle.net/1843/ESBF-8ZEN86

This work addresses the problem of decentralized navigation and control of teams of cooperative robots in cluttered environments. We propose here a methodology for the coordination of heterogeneous agents subject to disturbances and uncertainties, typical of complex environments. Our method is based on an approach of path and trajectory planning with random exploration of spaces, whose advantages of efficiency and generalization allow its application in real-time multirobot problems. We explore two mzin features of the approach: i) the sharing of a pseudo-random seed between teammates and ii) the usage of abstractions (parametric functions) for coordinated planning. At first, the robots use deterministic functions, that simulate randomized characteristics, to generate plans that are the same as of other robots onthe basis of the real-time methodologies. This allows maintaining the advantages of these techniques in a problem with deterministic solution. Secondly, the robots use parametric functions that make it possible the planning of the team in a distributed manner, without requiring that each agent be aware of the existence of others. That contributes for the scalability of the method, making it decentralized. The proposedmethodology is applied to two kinds of complex problems: i) formation control of a group of aerial robots and ii) air-ground cooperation of vehicles. We use as basis of prediction, models that incorporate both the dynamic and the kinematic behavior of the robots, as well as the main constraints in the state space. The controllers employed in this approach are simple, but the method is general enough to incorporate more sophisticated control laws