Ferramentas de auxílio ao desenvolvimento de preditores e controladores neurais completos a horizonte fixo para robôs móveis multiarticulados

• Ano de defesa: 2011
• Autor(a) principal: Miranda, Victor Marques
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Robótica; Robôs móveis; Robôs - Sistemas de controle; Sistemas de controle inteligente; Redes neurais (Computação); Interface gráfica com o usuário (Sistemas de computação); Engenharia Elétrica; 621.3
• Link de acesso: http://repositorio.ufes.br/handle/10/9620

The robotic navigation is a subject that has motivated several academic and practical works. Among all the different kinds of mobile robots, the multi-articulated one is the focus of this work. This dissertation presents new systematic and new tools for the development of full neural predictors and controllers, with fixed time horizon, based on static multilayer feedforward networks, when describing backward movements of multi-articulated mobile robots, in the configuration space. The predictors are necessary for robot’s assisted tasks and useful to be used as cores in simulators to analyze navigation strategies and for controller’s synthesis and validation. The proposed systematic and the developed tools are general. The implemented tool helps to make easier the processes of creating, training and validating several topologies of neural networks and setting their training parameters. Besides, this tool helps to find a prediction horizon, for a given robot, regarding the fact that it is an exhaustive task. At first, the characteristics of the prototype used in this work, similar to a real vehicle, keeping the proper scales, and its kinematic chain are presented. Then, the robot is modeled using static neural networks with different prediction horizons. It is established a strategy for the data acquisition in order to obtain a representative database that will be used for training and validation of the models. It is implemented a preprocessing step, where some manipulations are done on data until they can reach an appropriate format for predictors training. The training data set is composed by real data acquired from measurements of the prototype and by data generated from circular singular conditions models. This work presents models for the singularities and for the critical angles extracted from an original analytical model of general backward movement equations of a multi-articulated mobile robot, in the configuration space. The same model is used to generate analytical predictors, which are used with the neural controllers. The use of models for singularities is necessary because the singular conditions are situations of unstable equilibrium, which makes impossible to obtain enough data from open loop real systems. The model for critical angles defines the range of the configuration variables before the jackknife. Finally, it is shown the generation of inverse controllers directly from the collected data and from the singular models or indirectly from the predictor, in the configuration space. The validation of the controller aims to reveal its capacity of following the references within an acceptable horizon and error, avoiding jackknife situations and keeping the convexity and the circular movement.