Adaptive BCI-Controller For Dynamic Systems

of the Thesis Adaptive BCI-Controller For Dynamic Systems by Golnaz Eftekhari Yazdi Master of Science in Electrical and Computer Engineering Northeastern University, July 2015 Dr. Deniz Erdogmus, Adviser This thesis introduces a new approach to design an adaptive proportional controller for dynamic systems controlled by Brain Computer Interfaces (BCIs). Brain-controlled robots are being developed to help people with severe disabilities in their daily lives. The majority of work in this field considers the BCI output as a discrete value that represents the user decision or intent. In these studies the reaction of the dynamic system being controlled to user decisions estimated with different levels of confidence is the same. Therefore user mistakes or misclassification errors impact overall system behavior considerably. One way to address this issue is to impose a minimum confidence threshold for decisions generated by the BCI and conveyed to the system, but this approach slows down the decision making rate. In this study, BCI output is a continuous valued variable related to confidence level of the instantaneous decisions by the BCI regarding user intent. The proposed method characterizes the error behavior of BCI inference based on electroencephalogram (EEG) evidence, and exploits these statistical properties of the BCI classifier. A simple adaptive controller that conveys BCI outputs to the dynamic system being controlled utilizes is optimized recursively using these statistical properties and any prior knowledge of user intent. The performance of the proposed BCI-controller has been estimated by computer simulations based on real Steady State Visually Evoked Potential (SSVEP) datasets. According to these