Development of alternative kinematic solutions for Cartesian control of a Series Elastic Actuated upper limb exoskeleton.

Robot-assisted rehabilitation is an arising approach used to help post-stroke patients to recover motor capacity. A stroke occurs when the carrying of oxygen and nutrients to the brain is interrupted and so part of the brain cannot get the blood it needs. The consequence is the death of brain cells in the region that controls a particular body function, as reflection a part of the body stops working as it is supposed to do. Stroke causes in patients a loss of hand dexterity and motor impairment on the upper-limb movements, such as reaching and grasping, which are fundamental components of many of the Activities of Daily Living (ADLs). Therefore, these limitations drastically constrain the health-related quality of life of the subject. Robotic mediated rehabilitation allows both to obtain a distributed interaction along the whole limb to be rehabilitated, and to perform movements with the same quality of the ones performed during the therapist mediated rehabilitation. The usage of sensors on the mechatronic structure allows to collect a relevant set of feedback data that can be used both to evaluate the performance of the exoskeleton from a robotic point of view, and to understand and improve the human-robot interaction. The first part of the thesis focuses on the modelling, validation and driving of a torque sensitive actuator that can be employed to improve the human-robot interaction. A model of the motor has been developed inside a simulation environment (Simulink and Matlab), and then characterized with the data available in the datasheet of the mechanical components. The system is a Series Elastic Actuator (SEA) composed by a DC brushless motor, a harmonic drive, and a compliant element that provides a measure of the exerted torque. Different models of motors with different performances and different springs have been analyzed. The model developed has been validated by comparing the simulated data with the one collected at the test bench, demonstrating that the simulation can be used for testing purpose instead of the physical test bench to save time and avoid a damage in the mechanical components. All the tests have been performed on a real-time target machine. During the second part of the thesis, it has been studied the kinematics of the shoulder joint complex of an upper limb active exoskeleton; after the study of the direct kinematics, an un-conventional method has been adopted to implement the inverse kinematics by employing a neuro-Fuzzy-Logic based technique. A first model has been deployed for off-line simulations, another model has been deployed to a real-time target machine (Speedgoat), then the results have been compared. The exoskeleton can follow a desired trajectory defined inside the cartesian space by exploiting the position control employed by the drivers, and the on-line computed neuro-Fuzzy-Logic based inverse kinematics. It has also been developed an admittance control that exploits the potentiality of the SEA to make the system more compliant during the machine-human interaction.