A large number of people with severe motor disabilities cannot access any of the available control inputs of current assistive products, which typically rely on residual motor functions. These patients are therefore unable to fully benefit from existent assistive technologies, including communication interfaces and assistive robotics. In this context, electroencephalography-based Brain-Machine Interfaces (BMIs) offer a potential non-invasive solution to exploit a non-muscular channel for communication and control of assistive robotic devices, such as a wheelchair, a telepresence robot, or a neuroprosthesis. Still, non-invasive BMIs currently suffer from limitations, such as lack of precision, robustness and comfort, which prevent their practical implementation in assistive technologies. The goal of this PhD research is to produce scientific and technical developments to advance the state of the art of assistive interfaces and service robotics based on BMI paradigms. Two main research paths to the design of effective control strategies were considered in this project. The first one is the design of hybrid systems, based on the combination of the BMI together with gaze control, which is a long-lasting motor function in many paralyzed patients. Such approach allows to increase the degrees of freedom available for the control. The second approach consists in the inclusion of adaptive techniques into the BMI design. This allows to transform robotic tools and devices into active assistants able to co-evolve with the user, and learn new rules of behavior to solve tasks, rather than passively executing external commands. Following these strategies, the contributions of this work can be categorized based on the typology of mental signal exploited for the control. These include: 1) the use of active signals for the development and implementation of hybrid eyetracking and BMI control policies, for both communication and control of robotic systems; 2) the exploitation of passive mental processes to increase the adaptability of an autonomous controller to the user’s intention and psychophysiological state, in a reinforcement learning framework; 3) the integration of brain active and passive control signals, to achieve adaptation within the BMI architecture at the level of feature extraction and classification.
|Titolo della tesi:||Co-adaptive control strategies in assistive Brain-Machine Interfaces|
|Data di discussione:||6-feb-2018|
|Appare nelle tipologie:||Tesi di dottorato|