Towards A Semi-Autonomous Brain-Computer Interface

Brain-computer interfaces (BCI) provide severely disabled people with the means to control assistive devices. However, there is a high mental workload for many BCI users. In many systems, users must actively control each of the machine's low-level actions. Recent studies postulate an alternative approach, using spontaneously-generated brain signals, while users merely observe machine actions.

It is possible to differentiate the brain's responses to correct and incorrect machine actions, using single-trial electroencephalography (EEG). Furthermore, having classified actions as correct or erroneous, robots can use machine learning to find quasi-optimal routes to a target. A few studies have differentiated the brain's responses to errors of different direction or severity. However, errors cannot always be categorised in these ways.

This thesis firstly shows that it is possible to differentiate, using single-trial EEG, subtly different errors that could not be distinguished by existing metrics: stepping off a target, or moving further from the target when starting from an off-target location. An additional data set is used to further validate the feasibility of distinguishing different error types. This thesis then shows for the first time that it is possible to distinguish EEG responses to different correct navigational actions, gaining specific information indicating when a target has been reached. Finally, a system is presented which responds to the detailed classification of these navigational actions in real-time. [The details of this navigational system are redacted in the public version of this thesis due to a temporary embargo. See Chapter 5 for further details of the embargo.] This novel strategy facilitates more efficient virtual robot navigation and target identification than current state-of-the-art approaches.

In bringing these advances together, this thesis presents the foundation of a new framework for detailed implicit brain-machine communication. This facilitates semi-autonomous robot control, reducing the mental burden for BCI users.