Improving classification of error related potentials using novel feature extraction and classification algorithms for an assistive robotic device

We evaluated the proposed feature extraction algorithm and the classifier, and we showed
that the performance surpassed the state of the art algorithms in error detection. Advances in
technology are required to improve the quality of life of a person with a severe disability who
has lost their independence of movement in their daily life. Brain-computer interface (BCI)
is a possible technology to re-establish a sense of independence for the person with a severe
disability through direct communication between the brain and an electronic device. To enhance
the symbiotic interface between the person and BCI its accuracy and robustness should
be improved across all age groups. This thesis aims to address the above-mentioned issue by
developing a novel feature extraction algorithm and a novel classification algorithm for the
detection of erroneous actions made by either human or BCI. The research approach evaluated
the state of the art error detection classifier using data from two different age groups, young
and elderly. The performance showed a statistical difference between the aforementioned age
groups; therefore, there needs to be an improvement in error detection and classification. The
results showed that my proposed relative peak feature (RPF) and adaptive decision surface
(ADS) classifier outperformed the state of the art algorithms in detecting errors using EEG for
both elderly and young groups. In addition, the novel classification algorithm has been applied
to motor imagery to improve the detection of when a person imagines moving a limb. Finally,
this thesis takes a brief look at object recognition for a shared control task of identifying utensils
in cooperation with a prosthetic robotic hand.