Assistive devices can greatly enhance the quality of life of individuals with disabilities. Based on the user’s ability and preference different modalities can be used to control these assistive devices, such as existing muscle movements, voice, eye gaze, and brain signals. However, despite recent advances in technology, most of the assistive devices used by people with severe motor disabilities have limited and very slow task performance, since a slight increase in the complexity of the system extensively increases the required inputs from the user resulting in intolerable physical and mental workload.
This study aims to address this challenge by introducing implicit interactions between a human and an assistive device through brain signals. In such a system a considerable part of inputs required for controlling the assistive device is derived from automatic, spontaneous brain activity, interpreted in the given context. Importantly, the user exerts no effort to actively, explicitly, or voluntarily elicit or modulate this brain activity. Instead, the user focuses on the task at hand while a passive brain-computer interface (BCI) system, in the background, monitors their brain activity for informative correlates of relevant mental states (e.g. erroneous actions, target selection).
In this study a virtual robot had to navigate towards, and identify,target locations in both small and large grids, in which any location could be the target. For the first time, we apply a system utilising detailed EEG feedback: 4-way classification of different types of movement is performed, including specific information regarding when the target has been reached. Additionally, we perform binary classification of whether targets are correctly identified. Our proposed strategy implements Bayesian inference to infer the most likely target location from the brain’s responses.The experimental results show that our novel use of such detailed feedback facilitates a more efficient and robust system than the state-of-the-art. Furthermore, unlike state-of-the-art approaches,the results show scalability of our proposed approach: By tuning parameters appropriately, our strategy correctly identifies over 98% of targets, even in large search spaces