A learning-based model-free controller for decoupled humanoid robot walking
Carregando...
Data
2020-12-30
Autores
Tejada Begazo, Maria Fernanda
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
A robótica avançou rapidamente nas últimas décadas. Inicialmente focada na produção industrial de automóveis depois da segunda revolução industrial; ela rapidamente se expandiu para novos e excitantes domínios que demandaram dos robôs a realização de tarefas com ferramentas e em ambientes projetados para seres humanos. No entanto, controlar o movimentos desses robôs está se tornando cada vez mais desafiador devido ao crescimento da complexidade de sua estrutura mecânica e dinâmica. Essa dificuldade tem estimulado a investigação de novos métodos para o controle de robôs que possam não depender do modelo cinemático ou dinâmico dos robôs. A robô humanóide Marta foi projetada e construída pelos grupos de pesquisa LaRoCS e GASI para a realização de investigações de caminhada de robôs humanóides, e apresenta alguns novos conceitos tais como os pés pequenos - ao invés de pés grandes - e uma junta esférica de 3 DOF na cintura que permite o projeto de estratégias de controle desacopladas. Aproveitando a arquitetura mecânica deste robô, este trabalho apresenta um controlador livre de modelo para a caminhada de robôs humanóides composta por i) um controlador clássico para a junta esférica da cintura que foca na estabilidade do robô e ii) um controlador de trajetória para as demais juntas baseado em uma Série Truncada de Fourier (TFS) com seus parâmetros ajustados por um algoritmo genético (GA) que foca na implementação de um padrão de caminhada coordenada do robô. Os resultados mostraram que a estratégia de controle descomposta tem potencial para conferir robustez adicional à caminhada de robôs bípedes em relação a técnicas convencionais.
Robotics has advanced rapidly in recent decades. Initially, it focused on automotive industry production after the second industrial revolution; it rapidly expanded to new exciting domains that prompted robots to perform tasks with tools and in environments designed for humans. However, controlling the movements of such robots is becoming more and more challenging due to the growing complexity of their mechanical structure and dynamics. This difficulty has stimulated the investigation of novel methods for robots controlling that may not depend on the kinematic or dynamic models of the robots. The Marta humanoid robot was designed and built by the LaRoCS and GASI research groups for humanoid walking investigation and present some novel concepts like small feet – instead of large ones – and a 3-DOF spherical joint in the waist that allows the design of decoupled controlling strategies. Taking advantage of the mechanical architecture of this robot, this work presents a model-free controller for humanoid robot walking composed by i) a classical controller for the waist spherical joint that focuses on robot stability, and ii) a trajectory controller for the remaining joints based on a Truncated Fourier Series (TFS) its parameters tuned by a Genetic Algorithm (GA) that focus on the implementation of a coordinated robot walking pattern. Results have shown that the decoupled robot control strategy has the potential to confer additional robustness to the bipedal robots walking while compared to traditional techniques.
Robotics has advanced rapidly in recent decades. Initially, it focused on automotive industry production after the second industrial revolution; it rapidly expanded to new exciting domains that prompted robots to perform tasks with tools and in environments designed for humans. However, controlling the movements of such robots is becoming more and more challenging due to the growing complexity of their mechanical structure and dynamics. This difficulty has stimulated the investigation of novel methods for robots controlling that may not depend on the kinematic or dynamic models of the robots. The Marta humanoid robot was designed and built by the LaRoCS and GASI research groups for humanoid walking investigation and present some novel concepts like small feet – instead of large ones – and a 3-DOF spherical joint in the waist that allows the design of decoupled controlling strategies. Taking advantage of the mechanical architecture of this robot, this work presents a model-free controller for humanoid robot walking composed by i) a classical controller for the waist spherical joint that focuses on robot stability, and ii) a trajectory controller for the remaining joints based on a Truncated Fourier Series (TFS) its parameters tuned by a Genetic Algorithm (GA) that focus on the implementation of a coordinated robot walking pattern. Results have shown that the decoupled robot control strategy has the potential to confer additional robustness to the bipedal robots walking while compared to traditional techniques.
Descrição
Palavras-chave
Androids, Artificial intelligence, Genetic algorithms, Intelligent control systems