Contributions à l'exploitation d'exosquelettes actifs pour la rééducation neuromotrice

Abstract

Neuro-motor rehabilitation is one the the new application areas of physically interactive robots. In this domain, the aim is to design machines that are able of assisting an impaired patient's motions when he/she practices physical exercises. One of the key topics is to provide machines capable of finely mastering mechanical forces distributed along human members. This has motivated the development of robotic exoskeletons. However, most of the research so far has focused on kinematics of these devices, with little attention paid to the force transmission question. Rather, this thesis is mostly dedicated to this crucial aspect. The main contributions are in the domain of design and control of robotic exoskeletons, with a main objective defined by the quality of force control. In the domain of design, we have exploited literature results regarding the robot structure and actuation mechatronics. Indeed, a starting point of this research is an existing back-drivable exoskeleton called ABLE, designed by CEA-LIST for assisting human arm. Given this particular robot, which is representative of the state of the art in terms of mechanical properties, we have worked on the mechanical coupling between the robot and the human arm. This has yielded to a general method aimed at designing passive fixation mechanisms between exoskeletons and human members. These mechanisms allows to guarantee global isostaticity and to select transmitted forces in such a way that they are fully controllable. The method is successfully applied to ABLE, with experimental evidence of an increased quality of interaction. Regarding the control aspects, our work has focused on the so-called transparent mode, when the robot has to follow the patient's movement while minimizing the disturbances. We have deployed an original multi-contact force controller, which also uses as an input a prediction of the patient's movement. This mixed force/position controller is, again, successfully experimented on ABLE. A last result of this work concerns evaluation of the interaction quality during comanipulation tasks. Indeed, in order to be able of quantifying improvements brought by design and control developments, it has been necessary to setup a method aimed at reproducibly evaluate the quality of human-robot physical interaction. This is achieved through analyzing simultaneous records of forces and gesture kinematics.