Classification of error-related potentials evoked during observation of human motion sequences

Abstract :

In recent studies, electroencephalogram (EEG)-based interfaces that enable to infer human intentions and to detect implicit human evaluation contributed to the development of effective adaptive human-machine interfaces. In this paper, we propose an approach to allow systems to adapt based on implicit human evaluation which can be extracted by using EEGs. In our study, human motion segments are evaluated according to an EEG-based interface. The goal of the presented study is to recognize incorrect motion segments before the motion sequence is completed. This is relevant for early system adaptation or correction. To this end, we recorded EEG data of 10 subjects while they observed human motion sequences. Error-related potentials (ErrPs) are used to recognize observed erroneous human motion. We trained an EEG classifer (i.e., ErrP decoder) that detects erroneous motion segments as part of motion sequences. We achieved a high classification performance, i.e., a mean balanced accuracy of 91% across all subjects. The results show that it is feasible to distinguish between correct and incorrect human motion sequences based on the current intentions of an observer. Further, it is feasible to detect incorrect motion segments in human motion sequences by using ErrPs (i.e., implicit human evaluations) before a motion sequence is completed. This is possible in real time and especially before human motion sequences are completed. Therefore, our results are relevant for human-robot interaction tasks, e.g., in which model adaptation of motion prediction is necessary before the motion sequence is completed

Files:

Kim_Paper_HaLeR.pdf

Links:

https://link.springer.com/chapter/10.1007/978-3-031-35017-7_10