Social Impressions of the NAO Robot and its Impact on Physiology

TL;DR

This study investigates how NAO robot voice and motion modalities shape social impressions and concomitant physiological responses in neurotypical adults. It combines perceptual surveys (Godspeed/RoSAS) across five features with BVP-based physiological data, analyzed via two deep-learning pipelines (CNN on raw BVP and CNN on Gramian Angular Field representations). Results show modality-induced differences in perceived safety, anthropomorphism, likeability, and animacy, while perceived intelligence remains less affected; raw BVP with CNN achieves notably higher classification accuracy ($0.8179$) than the GAF-based approach ($\sim0.59$). The work demonstrates the potential to tailor HRI interventions by modality and provides a groundwork for extending to ASD populations and richer physiological signals, while noting limitations and future directions (e.g., candid dialogue, denoising, cross-population generalization).

Abstract

The social applications of robots possess intrinsic challenges with respect to social paradigms and heterogeneity of different groups. These challenges can be in the form of social acceptability, anthropomorphism, likeability, past experiences with robots etc. In this paper, we have considered a group of neurotypical adults to describe how different voices and motion types of the NAO robot can have effect on the perceived safety, anthropomorphism, likeability, animacy, and perceived intelligence of the robot. In addition, prior robot experience has also been taken into consideration to perform this analysis using a one-way Analysis of Variance (ANOVA). Further, we also demonstrate that these different modalities instigate different physiological responses in the person. This classification has been done using two different deep learning approaches, 1) Convolutional Neural Network (CNN), and 2) Gramian Angular Fields on the Blood Volume Pulse (BVP) data recorded. Both of these approaches achieve better than chance accuracy 25% for a 4 class classification.

Figures (7)

• Figure 1: Social impressions of the NAO robot during HRI and its impact on physiology. Statistical Inference Block evaluates the importance of different modalities of the robot with respect to perceived features (pf). The Deep Learning block presents a time series classification approach to differentiate between the difference in physiology during different conditions A-D.
• Figure 2: Deep learning models used for the two approaches. Figure \ref{['fig:raw_signal_cnn']} shows the model input of the raw signals. Figure \ref{['fig:GAF_CNN']} shows the use of Gramian Angular Fields with the CNN model.
• Figure 3: Test for equal variance in the populations for conditions A, B, C, and D each for perceived safety, anthropomorphism, likeability, animacy, and perceived intelligence.
• Figure 4: One-way ANOVA for finding significance of conditions A-D for different perceived robot features.
• Figure 5: Effect of experience with robotics on different perceived features under different conditions.