Single trial recognition of error-related potentials during observation of robot operation

Recent works suggest that several human cognitive processes elicited during the observation and monitoring of tasks developed by others can be detected in real time. These works have also demonstrated that human brain activity can be used to recover from machine errors, and as reward signals to teach a simulated robot how to perform given tasks. This paper studies the elicitation of this activity during the operation of a real robot. Experimental results have been obtained with 4 participants observing the operation of a 5 d.o.f. robotic arm performing correct/incorrect reaching tasks, while an EEG system recorded their brain activity. The results give evidence that the brain areas that play a role in detection and monitoring of errors also play a role when observing the operation of a real robot, that a brain discriminative response is elicited during the observation of a correct/incorrect operation of a real robot, and that it is possible to learn a classifier that provides online categorization with high accuracy (80%).