Stochastic model predictive control for dynamic locomotionof legged robots

• Ano de defesa: 2022
• Autor(a) principal: Schulze, Lucas
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Engenharia elétrica; Robôs; Controle preditivo
• Link de acesso: https://repositorio.udesc.br/handle/UDESC/15905

Legged robots are considered hybrid dynamics systems due to the presence of continuous anddiscrete dynamics resultant of the switching of the footholds. Especially on rough terrains, adesired motion plan may not be feasible, while a controller aims to execute it maintaining therobot in equilibrium. In this context, Model Predictive Control (MPC) is gaining popularity dueto the possibility to consider constraints and future changes on the footholds and in the terrain.However, to formulate computational tractable optimization problems, simplified mathematicalmodel are employed, leading to system uncertainties that may affect the execution of the motionplan and the robot’s stability. Stochastic Model Predictive Control (SMPC) emerged to addressthe presence of uncertainties in the system description into the controller design, balancingconstraint satisfaction and conservatism. Thus, this work proposes the comparison of MPC andSMPC for legged locomotion on rough terrains, and the evaluation of the effect of the angularmomentum rate minimization. MPC and SMPC are compared in a simulation scenario withobstacles for the HyQ robot. As result, SMPC and the minimization of the angular momentumrate demonstrated to improve the legged robot locomotion with relation to constraints satisfactionand trajectory smoothness