Título: Aproximação da Cinemática Inversa de um Robô Manipulador de 5 Juntas Rotativas através de Redes Neurais Globais e Locais
Autores: José Matheus Soares Ferreira, Icaro Bezerra Viana & David Nascimento Coelho
Resumo: The inverse kinematics calculation of robotic manipulators with multiple degrees of freedom is a challenge, where analytical methods are often not feasible. In this work, the approximation of the inverse kinematics of a robotic manipulator with five revolute joints is performed using two machine learning approaches: global models and local models, both based on Multi Layer Perceptron (MLP) neural networks. Initially, a dataset relating the joint and Cartesian spaces is generated through forward kinematics. Then, this dataset is filtered to remove redundant and unreachable samples for the manipulator. Finally, the models hyperparameters are optimized, and their performance is evaluated using the Mean Squared Error (MSE) in both the joint and Cartesian spaces. The results show that both approaches are viable for the given task, with the local approach outperforming the global one.
Palavras-chave: Robotics Manipulators; Multilayer Perceptron; Inverse Kinematics; Local Neural Network; Global Neural Network.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1191734
Artigo em PDF: CBIC_2025_paper1191734.pdf
Arquivo BibTeX:
CBIC_2025_1191734.bib