Following the Robot's Lead: Predicting Human and Robot Movement from EEG in a Motor Learning HRI Task

Abstract

Human-robot interaction (HRI) offers unique opportunities to study the neuroscience of human motor control through controlled and reproducible sensory stimuli. In this study, we introduce an innovative neuroscience-HRI framework inspired by the Serial Reaction Time (SRT) task, that combines EEG with a task where a humanoid robot performs preprogrammed movement sequences that are mirrored by a human participant in real time. The use of a humanoid robot ensures precise and repeatable sensory-motor stimuli in the 3D peripersonal space of the participant, providing experimental conditions that may be challenging to replicate with traditional methods. Behavioral performance is assessed by measuring the temporal lag between human and robot movements, which decreases with training, reflecting motor sequence learning. Concurrently, EEG data from the human participant is analyzed to reveal neural correlates of learning and movement dynamics. Event-Related Spectral Perturbations (ERSP) in theta, mu, and beta frequency bands demonstrate distinct patterns associated with rest, fixation, and movement. Furthermore, the ERSP changes over successive trials reflect the progression of sequence learning, highlighting the relationship between neural oscillations and motor learning. A Markov-Switching Linear Regression model further decodes EEG signals to predict movement parameters including both human and robot position and velocity in a time-resolved manner. Our findings highlight the potential of HRI as a robust platform for neuroscience research and underscore the value of EEG-based neural decoding in understanding motor sequence learning. This work suggests further advances for integrating robotics into neuroscience and rehabilitation research.