REDE NEURAL APLICADA Modelagem e Controle em Robôs Quadrúpedes

The application of Neural Networks is fundamental for pattern identification, correlation establishment, classification, and result prediction. This study leverages these capabilities to develop a Mobile Robotic Embedded System (MRES) consisting of two Neural Networks. The first acts as a 'model,' training the controller to operate predictively, while the second serves as the controller. The goal is to create an adaptable neural controller capable of learning from the environment through collected sensory data. This controller adjusts its actions based on this information, ensuring responses meet established performance criteria. Beyond the implementation of neural models, this study investigates an integrated approach that combines prediction and control techniques, exploring the potential of neural networks in a robotics context. The success observed in the proposed strategies suggests prospects for future research, considering various neural network architectures and refinements in system dynamics. The proposed control system stands out for its adaptability, continuous learning capability, and operational efficiency in environments characterized by challenging dynamics, paving the way for the development of new controller paradigms.