A Brain-Computer Interface for robotic arm control

Abstract

Brain-Computer Interfaces (BCI) are tools that open a new channel of communication between humans and machines. \nThe majority of human input devices for computers require proper functioning of our primary sensors and motor \nfunctions like grasping, moving and visual perception. In the case of severe motor disabilities, like amyotrophic \nlateral sclerosis (ALS) or spinal chord injury (SCI),\nThe most common method to measure brain activity suitable for BCI are electroencephalographic measurements (EEG) \ndue to their relative cost effectiveness and ease of use. Alternative ways to extract brain signals exist but \neither require invasive procedures, i.e. opening the skull, or are very costly and bulky (MEG, fMRI) \nwhich renders them unusable for home appliance. \nOne of the most popular brain controlled input methods is the P300-Speller paradigm\nwhich gives the user control over a virtual keyboard to enter text. The term P300 refers to a specific EEG component \nthat can be measured whenever a rare task relevant stimulus is interspersed with many non-relevant stimuli. This method requires \nthe ability to control the visual presentation of stimuli and therefore also requires some sort of computer controlled display. \nThe recognition rates for this type of BCI, yet already quite high with roughly 80-90% accuracy, \nare still prone to errors and may not be suitable for critical applications like issuing movement commands to a wheelchair \nin a highly populated environment. Commands to stop the wheelchair might be recognized too late. \nFurther, it is impossible with the standard stimulus matrix to react to external influences like obstacles or select physical objects \nin a scene which does not allow the user to interact with a dynamic environment. \nThis work aims to fuse state of the art BCI techniques into one single system to control an artificial actuator \nlike a robot arm and use it to manipulate the physical environment. To achieve this goal, multiple techniques originating \nfrom different fields of research as augmented reality, computer vision, psychology, machine learning and data mining have \nto be combined to form a robust, intuitively to use input device.