Patient-tailored bidirectional interfaces for rehabilitation and assistive robots

Bidirectional interfaces, which combine recording and stimulating systems in so-called closed-loop devices, are the new generation of interfaces. In order to guarantee attentive usage and fine control of the device, they are typically tailored on the users' particular needs and are designed by taking into account users' residual physical and cognitive abilities. Ambition of this thesis is to design and develop a patient-tailored bidirectional interface for rehabilitation and assistive robots that i) is adaptable to the user's residual functional and motor capabilities and ii) works well in unstructured environments and with different robot types (e.g. manipulators, exoskeletons or prostheses). The developed bidirectional interface is composed of two main modules, namely the humanmachine interface for the device control and the interface for sensory feedback. The human-machine interface for the device control was designed to be used by patients with different level of disabilities to drive their rehabilitation or assistive robot, e.g. an upper-limb prosthesis, a robotic manipulator or an upper-limb exoskeleton, both continuously and by means of a trigger-based approach. The interface proposed in this work is based on the coupled use of myoelectric and magnetoinertial sensors. It was first designed to be used by trans-humeral amputees to control their prosthetic device. In particular, with the proposed approach the user could operate the elbow flexion-extension, wrist prono-supination and hand opening-closing by exploiting the residual stump motions combined to the myoelectric activity of two target muscles, i.e. biceps and triceps. The proposed control interface was tested by eight healthy subjects who were asked to drive a trans-humeral prostesis in a virtual environment. A comparative analysis between the proposed control and the traditional myoelectric control used in literature to drive commercially available prostheses was carried out. Results demonstrated that the user, by using the proposed method, could manage simultaneous movements and more physiological reaching tasks compared to the traditional myoelectric control that enables only sequential movements. Subsequently, adaptability of the proposed control interface to patients with different levels of disability and different robot types were demonstrated. It was tested on people with severe motor disability to control their robotic rehabilitation/assistive device, such as a manipulator or an upper-limb exoskeleton, both continuously and by means of a trigger-based approach. Two experimental sessions were carried out. The first experimental session was aimed to compare the proposed interface, based on magneto inertial sensors and myoelectric electrodes, to a standard interface made of the voice recognizer. Sixteen healthy subjects were asked to continuously control the motion of a robotic manipulator, by using the two control interfaces, for assistive purposes. The obtained results pointed out that performance and level of acceptance were higher for the proposed interface with respect to the voice control. The second experimental session was aimed to evaluate user's preferences related to the amount of his/her intervention in the robot control. Two control modalities were implemented in order to modulate the frequency of the user's intervention in the robot control depending on the user's cognitive/physical state. They are the continuous control and the trigger control. They were compared in terms of effectiveness of the task fulfillment and user’s personal feelings related to the interface usage. The obtained results demonstrated a high patient involvement in using the continuous control, but better performance, in terms of effectiveness of the task fulfillment, has been achieved with the trigger-based control. Differently from the continuous control, the trigger-based control requires only a few actions to the user in order to start the robot movement. Hence, a motion planning system was developed in order to allow the robot autonomously accomplish the task in a way that is completely safe and accepted by the user. In this work a motion planning system for rehabilitation and assistive robotics, grounded on a Learning by Demontration (LbD) approach, was proposed. The LbD algorithm presented in this work is grounded on Dynamic Movement Primitives (DMPs), but it is improved in terms of i) accuracy of the trajectory reconstruction, ii) adaptability of the DMPs to different subjects' anthropometry and robotic devices (e.g. manipulators or exoskeletons) ii) ability to reproduce human-like movements, iii) ability to solve orientation singularity in the DMP equations and iv) generalization capabilities with respect to different target positions. This was confirmed by four experimental sessions that were carried out in order to assess the motion planner performance. The experiments involved healthy subjects and patients with Limb girdle muscular dystrophys who were asked to perform activities of daily living with the aid of different robot types, i.e. robotic manipulators and upper-limb exoskeletons. The interface for sensory feedback was designed and developed to improve user postural control during robot-aided daily living activities, both in standing and in sitting position. In particular, the proposed vibrotactile stimulation feedback was employed, during robotaided rehabilitation, to correct patients' spine posture. Three inertial sensors were used in order to measure trunk and neck flexion/extension (F/E) and information about user's incorrect posture were provided by two lightweight vibrating actuators located on the user's arms. The proposed stimulation feedback was compared to a typical approach used in literature to warn users about incorrect posture, i.e. visual feedback, in terms of i) effectiveness to improve the posture of the subject, ii) acceptability and iii) user's comfort. Ten healthy subjects were asked to perform 3D reaching movements with the aid of a robotic manipulator attached to their right wrist. During the rehabilitation session, they were provided with visual and vibrotactile feedback to retain their trunk and neck in a correct posture. Additionally they were asked to perform the tasks without any information about the correctness of their posture. The obtained results demonstrated that the users had a significant improvement in the spine posture when the task was performed with the aid of the visual and vibrotactile feedback compared to a no feedback condition.