Title : Single-trial classification of the P 300 event-related potential for shaping human-robot interaction

A. Finke,Yaochu Jin,H. Ritter

2009-01-01

Abstract:Brain-computer interfaces (BCI) basing on the P300 event-related potential (ERP) have been developed and studied widely in the past decade. As all ERPs, the P300 is a component in the human electro-encephalogram (EEG) defined by a specific polarity and a latency of occurrence after the onset of a stimulus, i.e. positivity and a latency of 300ms. These BCIs exploit the oddball paradigm, a well-studied method where stimuli are presented on a computer screen in an arbitrary fashion, to implement a binary selection mechanism used to control whichever device. We hypothesize that that the P300 is related to human surprise, meaning that it is generally triggered by rare and unpredictable events. The hypothesis is derived from the oddball paradigm where the oddballs – also rare and unpredictable events – evoke surprise in the human on a basic level. In human-robot interaction (HRI) a flawless behavior of the robot is needed in order to avoid puzzlement and/ or anxiety on the side of the human interaction partner. Detecting such reactions/ states in the human on the fly and providing instantaneous feedback during interaction is therefore crucial. Ideally, such a feedback mechanism does not demand additional cognitive load but operates “silently/unnoticed” in the background, i.e. it is an inherent feature of the human-machine interaction framework. Information extracted from the human EEG, in particular the P300 potential, is well suited for providing input to this feedback loop. Therefore, we propose to use P300 as a means for human-robot interaction, in particular to detect the surprise of a human being during interaction to detect in time any mistakes human observe so that the robot can notice this as early as possible to correct its mistakes in behavior. We propose a brain-robot interface that implements the aforementioned P300-based feedback approach. The core parts of the feedback system are (1) a “P300 spotter” that analyzes and classifies the incoming preprocessed data stream for P300 potentials on a single-trial basis and (2) a “translation” module that translates the detected P300s into appropriate feedback signals for the robot. It is important to note that no specific external triggers are needed to elicit the desired EEG components used for feedback generation. All feedback is generated according to components resulting inherently from the real interaction. The classification relies on a supervised machine learning algorithm that requires labeled training data. This data must be collected from the subject who wants to use the interface to account for the high inter-subject variances typically