Classifying human emotions in HRI: applying global optimization model to EEG brain signals

Area of Science:

• Human-Robot Interaction (HRI)
• Neuroscience
• Artificial Intelligence

Background:

• Social robots are increasingly designed to be more human-like for better acceptance.
• Brain-computer interfaces (BCIs) are being explored to enable robots to understand human emotions.
• Challenges in emotion recognition via BCIs include subjectivity, context dependency, and noisy data.

Purpose of the Study:

• To detect human emotional states from electroencephalogram (EEG) brain activity during human-robot interaction (HRI).
• To address challenges in real-time emotion recognition using BCIs.
• To improve the accuracy of classifying human mental states in response to robot behavior.

Main Methods:

• Collected EEG signals from 10 participants interacting with a Pepper robot exhibiting distinct personalities.
• Utilized emotion valence and arousal measures derived from frontal brain asymmetry (FBA).
• Trained machine learning models, optimized with a Global Optimization Model (GOM) for feature selection and hyperparameter tuning.

Main Results:

• Achieved classification accuracy of up to 92% for detecting user emotional responses from EEG signals.
• Demonstrated the feasibility of real-time emotion detection during human-robot interaction.
• The Global Optimization Model significantly improved classifier performance.

Conclusions:

• This study advances the Theory of Mind (ToM) in HRI by enabling robots to interpret user emotions.
• The findings pave the way for developing more empathetic and responsive social and assistive robots.
• Accurate emotion detection from EEG signals is crucial for more intuitive human-robot collaboration.