Conception et optimisation d’un robot de rééducation neuromotrice du membre supérieur avec compensation active de la gravité

In developed countries, the stroke is the third cause of adult mortality and involves disabilities with heavy consequences in daily living activities for a large part of survivors. Shortly after the attack, patients integrate a rehabilitation program to stimulate the recovery, at least partial, of its motor and cognitive skill. The lifestyle and demographic changes in developed countries lead to a continuous increase of the number of patients which require rehabilitation. This leads to increase the treatments efficiency or the research on new ways of treatments to improve patient recovery and decrease the care costs. The rehabilitation robots appear to be a novel tool to help therapists during rehabilitation tasks of the upper limb. Indeed, they permit, among others, to execute interactive tasks with high intensity in a motivating environment for the patient. However, some patients have a too weak neuromuscular system to execute rehabilitation exercises where their involvement in the movement generation is high. These patients are not able to hold their arm against gravity and can only undergo movements generated by the robot. Based on this observation, the main objective of this work is to develop an active gravity compensation within upper limb rehabilitation robots to help patients to execute high intensity active exercises earlier in the therapy and in a larger workspace. The method developed in this thesis consists in simulating in real-time an upper limb model to calculate the efforts due to gravity and to substract them in the control scheme of the robot. In this way, the arm weight is compensated through the robot actuators. This work is composed of three main parts: the development of an upper limb tracking system, the development of a robot for the rehabilitation of the shoulder complex and the development of the gravity compensation control strategy. The motion tracking system permits the estimation of the upper limb kinematics to align in real-time the upper limb model on the measured one. To achieve this, several calibration procedures and kinematics calculations have been implemented with the use of inertial and magnetic sensors. The developed system performances meet the specific needs: the sensors are lightweight, wearable and are able to estimate the proximal kinematics of the upper limb with an accuracy of 5°. The rehabilitation robot generates adequate efforts in patient joints to counterbalance arm weight and to help the patient to accomplish functional tasks during therapy. For this purpose, a novel alignment-free exoskeleton was developed for the rehabilitation of the shoulder complex. This robot is constituted of two actuated joints and is linked to the patient via three passive mechanical joints to manage two of the three rotational degrees of freedom of the shoulder. Its mechanical structure, dimensions and positioning relative to the patient have been optimized to fulfil at best all requirements. The first prototype of this exoskeleton named AFREXOS was materialized and an impedance-based controller was implemented to control arm's movements and interaction forces in the patient's articular coordinates. The presented characterization of the robot's performances shows that AFREXOS is able to mobilize the upper arm in a large part of the workspace of the shoulder with a maximal continuous shoulder torque of 39 Nm and a maximal speed of 120°/s. Moreover, the implemented impedance controller is able to simulate a free-motion with a very low impedance up to a frequency of 3.6 Hz. The developed compensation strategy permits to counterbalance precisely the upper limb weight in a large workspace. The trials carried out on healthy subjects showed that the implemented compensation control strategy reduces significantly the muscular activity necessary to hold the upper limb in several configurations where the gravity has the greatest impact on the articulation efforts.