Myoelectric Control Architectures to Drive Upper Limb Exoskeletons

Myoelectric interfaces are sensing devices based on electromyography (EMG) able to read the electrical activity of motoneurons and muscles. These interfaces can be used to infer movement volition and to control assistive devices. Currently, these interfaces are widely used to control robotic prostheses for amputees, but their use could be beneficial even for people suffering from motor disabilities where the peripheral nervous system is intact and the impairment is only due to the muscles, e.g. muscular dystrophy, myopathies, or ageing. In combination with recent robotic orthoses and exoskeletons, myoelectric interfaces could dramatically improve these patients’ quality of life. Unfortunately, despite a wide plethora of methodologies has been proposed so far, a natural, intuitive, and reliable interface able to follow impaired subjects’ volition is still missing. The first contribution of this work is to provide a review of existing approaches. In this work we found that existing EMG-based control interfaces can be viewed as specific cases of a generic myoelectric control architecture composed by three distinct functional modules: a decoder to extract the movement intention from EMG signals, a controller to accomplish the desired motion through an actual command given to the actuators, and an adapter to connect them. The latter is responsible for translating the signal from decoder’s output to controller’s input domain and for modulating the level of provided assistance. We used this concept to analyse the case of study of linear regression decoders and an elbow exoskeleton. This thesis has the scientific objective to determine how these modules affect performance of EMG-driven exoskeletons and wearer’s fatigue. To experimentally test and compare myoelectric interfaces this work proposes: (1) a procedure to automatically tune the decoder module in order to equally compare or to normalize the decoder output among different sessions and subjects; (2) a procedure to automatically tune gravity compensation even for subjects suffering from severe disabilities, allowing them to perform the experimental tests; (3) a methodology to guide the impaired patients through the experimental session; (4) an evaluation procedure and metrics allowing statistically significant and unbiased comparison of different myoelectric interfaces. A further contribution of this work is the design of an experimental test bed composed by an elbow exoskeleton and by a software framework able to collect EMG signals and make them available to the exoskeleton’s actuators with minimal latency. Using this test bed, we were able to test different myoelectric interfaces based on our architecture, with different modules choices and tunings. We used linear regression decoders calibrated to predict the muscular torque, low-level controllers having torque or velocity as reference, and adapters consisting of a properly dimensioned gain or simple dynamic systems, such as an integrator or a mass-damping system. The results we obtained allow to conclude that EMG-based control is a viable technology to assist muscular weakness patients. Moreover, all the components of the myoelectric control architecture – decoder, adapter, controller, and their tuning – significantly affect the task-based performance measures we collect. Further investigations should be devoted to a methodology to automatically tune all the components, not the decoders only, and to the quantitative study of the effect the adapter has on the regulation of the assistance level and of the tradeoff between speed and accuracy.