This thesis focuses on the key control problems in soft robotics. A step-by-step approach is proposed, from robot modeling to dynamic control and motion planning. In addition to the theoretical works, extensive experiments and simulations are also conducted to illustrate the effectiveness of the proposed methods. The thesis at first extends the previous work of our team, in which the linear modeling and control framework is proposed based on the reduced order Finite Element Method model. We thoroughly discussed the usage and advantages of an improved model order reduction method that preserves the structure and stability of FEM model. This new modeling method enables the design of a disturbance observer-based feedback-feedforward controller. However, the proposed control scheme is effective but limited to a small range where the linear model is effective. To perform the dynamic control in the whole nonlinear workspace of soft robot, we extended the linear control scheme to a new linear parameter varying (LPV) control framework. The LPV model is developed using both reduced-order models obtained from FEM models and the data collected from the soft robots. The stability of the generalized controller is guaranteed with the Lyapunov stability theory and the controller design is formulated as an optimization problem under linear matrix inequalities(LMI). However, LPV model is based on linearized models. The feedforward control and motion planning tasks can not be achieved effectively with LPV models. To achieve higher performance for the control of soft robots, we also proposed corresponding solutions for these issues. A novel kinematic representation of soft robot configurations combining both model and collected data is proposed, as well as a direct inverse kinematics method based on measurement data. For the feedforward control of a soft robot that can not be obtained analytically, preliminary results about the functional representation of feed-forward control and feed-forward learning laws with stability guarantees are illustrated.
Examinateur : Mme Christine CHEVALLEREAU Directeur de recherches, CNRS, Nantes Rapporteurs : Jamal Daafouz Professeur Université de Lorraine Edouard Laroche Professeur Université de Strasbourg Co-directeurs de thèse : Thierry Marie Guerra Professeur Université Polytechnique Hauts-de-France Alexandre Kruszewski Professeur Centrale Lille Institutes Co-encadrant : Anhtu Nguyen MdC Université Polytechnique Hauts-de-France Invited member : Christian Duriez Directeur de recherche INRIA Lille
Thesis of the team DEFROST defended on 27/01/2023
français
English