Evolutionary design for robust self-adaptive control

• Ano de defesa: 2024
• Autor(a) principal: Morais, Gustavo Alves Prudencio de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: algoritmos evolutivos; autonomous systems; controle robusto; evolutionary algorithms; otimização uni e multiobjetivo; robust control; single and multiobjective optimization; sistemas autônomos
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/18/18153/tde-08102024-161622/

Dealing with parametric uncertainties in mobile autonomous systems is a critical challenge. The difficulties scale when self-driving systems are operating in unconfined spaces or with interaction with people and other autonomous agents. Robust controllers have emerged as efficient solutions for ensuring autonomous navigation in such scenarios. However, uncertainty matrices for these agents are typically established using algebraic methods, necessitating prior knowledge of system dynamics. Consequently, control system designers rely heavily on the accuracy of uncertain models to achieve optimal control performance. To address these limitations, this study proposes a robust recursive controller developed through evolutionary optimization. A self-adapted algorithm was designed to incorporate robust control in single and multiobjective scenarios. Additionally, a local search strategy for addressing multiobjective optimization challenges is introduced. This methodology can be applied to any established multiobjective evolutionary algorithm found in existing literature. The findings demonstrate that this combination of a modelbased controller and machine learning significantly enhances system effectiveness in terms of robustness, stability, and smoothness. Moreover, this approach offers a more adaptable and comprehensive solution for scenarios with uncertainties in vision-based control and for heavy vehicles with significant mass variation.