Brain-wave bio potentials based mobile robot control: wavelet-neural network pattern recognition approach

We show how a recently developed wavelet methodology could be useful for EOG state classification. The work focuses on using EOG and EMG signals to drive a robot and our approach is characterized by our emphasis on wavelet pattern recognition methods rather than on the experimenter's feedback training. In our experimental system, we used a control device activated by human EOG and EMG signals from Cyberlink/sup TM/ to manipulate a robot's direction of movement. The EOG signal was processed by wavelet transform technique to facilitate the detection of EMG artifacts and to produce fuzzy signals for a neural network. We were successful in controlling the robot direction without the use of the experimenter's hands. The current implementation requires great concentration by the experimenter, and fatigue easily affects the desired performance of the system. To overcome these limitations, we propose a new kind of pattern recognition technique. The approach produces adequate classification when applied to EOG continuous data. Besides classification, this approach gives us valuable information about the relationship between EOG and EMG signals. This can be used to effectively detect, separate and remove EOG artifacts from even contaminated EMG signals variants.