Deep learning for emotional alignment in human-robot interaction (ELECTRA)

When we are engrossed in a conversation, we emotionally align with our partner: we unconsciously mimic each other, coordinate our behaviours and synchronize positive displays of emotion.

This mutual, co-adaptive emotional alignment has been associated with low-level, involuntary mechanisms of empathy and our ability to understand others.

This tremendously important skill, which spontaneously develops in human-human interaction (HHI), is currently lacking in robots.

To address this drawback, the ELECTRA project will develop new methods that leverage mining HHI data to bootstrap emotional alignment in human-robot interaction (HRI).

In combination with data automatically extracted from HHI using tools such as motion capture and automatic video analysis, traditional and emerging deep learning techniques will be used to capture correlations of non-verbal behaviours that are a prerequisite for emotional alignment and transfer them to HRI.

This will be done across robotic platforms with different human-likeness and different interaction contexts.

ELECTRA will run for 48 months and will include researchers (1 PI, 2 co-PIs, one researcher and 1 PhD student) with expertise in social robotics and HRI, machine learning, and social psychology.

By succeeding, the project will deliver new machine learning-based methods for human-robot emotional alignment transferrable across different robotic platforms and interaction contexts, key for a successful integration of robots in our daily lives.