Aplicações de sistemas classificadores para robotica autonoma movel com aprendizado
AUTOR(ES)
Lubnen Name Moussi
DATA DE PUBLICAÇÃO
2002
RESUMO
Mobile robotics has as its fundamental goal to elaborate trajectories avoiding collisions, locating and reaching targets, power supply self sufficiency, objects transportation, etc. A solution providing autonomy to the robot for learning its behavior is contrasting with more classical approaches that require a previous model of the environment in which the robot is inserted. In the case in which is needed to put the robot in another environment, or in which the environment has its configuration varying with time, these models present a great deficiency, demanding that the designer recalculates the trajectory, or even making its use unviable. The use of learning procedures releases the designer of inserting detailed knowledge of the environment and gives to the robot the possibility of behaving well in different environments. This work is directed to the solution of the robot s real time learning problem concemed with how to move avoiding collisions. To avoid collisions is essential for the movement of a mobile robot and is part of its wider strategy, whatever could it be its objective. Related to the approach to solve this learning problem, this work investigates, in a virtual environment, the use of Neural Networks within classifier systems, a solution not found in the literature. The use of neural networks has the intention of giving more descriptive power to the classifier systems, substituting its binary rules, limited in terrns of its processing power, by a more powerful tooL !t s also made a simulation of the Classifier Systems in its conventional form, which provides a comparison reference for the Neural Networks Classifier Systems. An interesting result obtained is trajectory smoothing provided by the neural networks. A number of suggestions are presented for future research. To achieve the intended results it was required to elaborate a Simulator, which is also part of this work, utilizes a 2 dimensional environment and takes into account some of the characteristics of a real robot, the mini robót Khepera
ASSUNTO(S)
robotica redes neurais (computação) sistemas inteligentes de controle
ACESSO AO ARTIGO
http://libdigi.unicamp.br/document/?code=vtls000293684Documentos Relacionados
- Ambiente de robotica para aplicações educacionais com SuperLogo
- Abordagem para criação de linguagens específicas de domínio para robótica móvel
- Estimação em sistemas com restrições de igualdade e aplicações em robótica móvel e de reabilitação
- RoboEduc: uma metodologia de aprendizado com robótica educacional
- Mapeamento semântico com aprendizado estatístico relacional para representação de conhecimento em robótica móvel.