Programação de robôs por demonstração utilizando modelos não lineares autorregressivos

• Ano de defesa: 2018
• Autor(a) principal: Rafael Francisco dos Santos
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Engenharia elétrica; Robótica
• Link de acesso: http://hdl.handle.net/1843/RAOA-BC6GXH

Autonomous robots are machines that operate without human intervention, fully automatic, using sensors to perceive the environment and make decisions employing that information. Due to the amount and complexity of the data generated by the robots and the difficulty in encoding complex movements, it is necessary to create new techniques to satisfactorily program control policies. One technique that has proven to be promising is programming by demonstration, where the general idea is to extract an adequate control law from demonstrations of the tasks to be performed. The main advantage of programming by demonstration is the direct encoding, from a set of demonstrations, of a given task, what allows robot to be programed by non-specialist people. This work proposes a methodology for learning robot reaching motions from a set of demonstrations using Nonlinear AutoRegressive (NAR) polynomial models. Reaching motions are modeled as solutions to autonomous discrete-time nonlinear dynamical systems so that the movements started near the data of the demonstrations follow the trained trajectories and always reach and stop at the target. Since NAR models obtained using standard system identification techniques do not always adequately model the reaching motions, in this work is presented a method that uses a least-squares estimator with constraints to impose the location of fixed points in the model. With the imposition of new fixed points it is possible to change the location of the original fixed points of the model, thus xi xii allowing the learning of stable reaching motions. The method was compared with the state-of-the-art using a calligraphy movement library. The results show that the proposed method is a good alternative in the area when it is needed a good precision in the trajectories learned, and, at the same time, a low demand of the low level controller. The method was also evaluated in an actual mobile robot and an industrial manipulator with success.