博士後期課程公聴会 （Fernandez Vargas Jacobo）
日時
平成30年1月18日(金) 10:00-11:30
場所
フロンティア医工学センター B号棟 1階会議室
発表者
Fernandez Vargas Jacobo
題目
Motion Reconstruction System for Trans-Humeral Amputees
主査
中川誠司 教授
副査
下村義弘 教授，並木明夫 准教授，兪文偉 教授(指導教員)
要旨
Motion reconstruction is a technology that uses motion-related bio-signals, such as electroencephalography (EEG) or electromyography (EMG), to estimate the trajectory of the motion. This technology can be used, for example, to control prostheses for amputees. However, so far, the results obtained with non-invasive methods are inadequate for real applications. In addition, most of the studies rely on a motion tracking system to obtain the data for training.
In this work, aiming to establish an effective noninvasive motion reconstruction system for trans-humeral amputees, who could only provide limited motion-related bio-signals, I proposed an approach including the following key components: 1) using both EEG and around-shoulder EMG as the input signal source; 2) exploring the most appropriate system architecture, while making clear the role of EEG and EMG for different motions; 3) investigating the possibility of training methods for trans-humeral prosthesis users in a virtual reality world, instead of using the commonly used motion tracking devices.
As a result, it was shown that: 1) by combining both EEG and around-shoulder EMG, accuracy of the reconstructed motion is better than that of any of them separately; 2) the most accurate and most robust architecture is the one that uses a two-layer motion estimator, in which the first layer predicts the position with EEG and EMG separately, and then, use the prediction of each of them to calculate the final prediction; 3) the training system with virtual reality technology could achieve overall accuracy comparable to other methods, even without using any motion tracking system. The findings obtained could be used to further improve the accuracy of EEG-based interface systems.
In this work, aiming to establish an effective noninvasive motion reconstruction system for trans-humeral amputees, who could only provide limited motion-related bio-signals, I proposed an approach including the following key components: 1) using both EEG and around-shoulder EMG as the input signal source; 2) exploring the most appropriate system architecture, while making clear the role of EEG and EMG for different motions; 3) investigating the possibility of training methods for trans-humeral prosthesis users in a virtual reality world, instead of using the commonly used motion tracking devices.
As a result, it was shown that: 1) by combining both EEG and around-shoulder EMG, accuracy of the reconstructed motion is better than that of any of them separately; 2) the most accurate and most robust architecture is the one that uses a two-layer motion estimator, in which the first layer predicts the position with EEG and EMG separately, and then, use the prediction of each of them to calculate the final prediction; 3) the training system with virtual reality technology could achieve overall accuracy comparable to other methods, even without using any motion tracking system. The findings obtained could be used to further improve the accuracy of EEG-based interface systems.