Cognitive Adaptation for Human Computer Interaction
MetadataShow full item record
Recognition of user's mental engagement is imperative to the success of Human Machine Interaction (HMI). By conducting a series of human experiments, we obtain implicit information about the cognitive and affective state of the subject using passive BCI recordings. We highlight the drawbacks of using conventional workload metrics as indicators of human engagement and assert that motor and cognitive workloads be differentiated. We propose new indices for flow measurements and advocate the development of an adaptive system that ensures high mental engagement while utilizing the benefits of implicit training in robotic rehabilitation via two case studies. We conduct passive BCI recordings on 8 subjects while they guide a robot with variable haptic resistance through a maze. We propose a new set of features for workload assessment, i.e. relative C3-C4 at beta and sigma frequency bands and differential Fz-POz at theta and alpha bands. A clear distinction between motor and cognitive workload has been made. We conclude that an increase in motor workload does not necessarily increase the cognitive workload and vice versa. We do not use any of the conventional workload metrics for engagement identification and suggest a novel paradigm for the same. In another study, we instruct 8 subjects to play a game with varying difficulty levels while recording their passive EEG activity. We use power spectral density and coherence values features as a combination to develop a classifier that separates the 3 difficulty levels in the game with a mean accuracy of 82.46%. We propose the use of higher frequency bands in flow measurement, as opposed to traditional flow classification methods that use lower frequency bands for the same. We conclude that high beta and sigma bands are the most informative bands in flow classification. Our method reports an accuracy increase of 19.98% when compared to the ones using the lower frequency bands for flow measurement. Finally, we demonstrate the efficacy and practicality of an adaptive human-robot shared control system using the higher accuracy and real time capabilities of our system in conjunction with the benefits of higher mental engagement and implicit training in therapy. Using the workload and flow information obtained by our method, robotic rehabilitation parameters can be adjusted in accordance with the user's needs. These adjustment may be in the force levels, difficulty level of the task or the speed of the task. Our method would further ensure that these adjustments are optimal for therapy while maintain a state of flow, thus avowing the efficacy of this novel paradigm in adaptive human-machine interaction.