Behavioural Models of Risk-Taking in Human-Robot Tactile Interactions

Area of Science:

• Human-robot interaction
• Affective computing
• Physiological computing

Background:

• Tactile interaction significantly influences human behavior and social dynamics.
• Previous research indicated that tactile intensity with robots affects human risk-taking propensity.
• Understanding the physiological underpinnings of this interaction is crucial for designing effective social robots.

Purpose of the Study:

• To investigate the relationship between human risk-taking behavior, physiological responses, and tactile interaction intensity with a social robot.
• To develop and evaluate machine learning models for predicting risk-taking behavior during human-robot tactile interaction.
• To identify key physiological and behavioral indicators of risk processing in human-robot interactions.

Main Methods:

• Collected physiological sensor data from participants playing the Balloon Analogue Risk Task (BART) during human-robot tactile interaction.
• Utilized a mixed-effects model as a baseline for predicting risk-taking propensity from physiological measures.
• Employed machine learning techniques, specifically Support Vector Regression (SVR) and Multi-input Convolutional Multihead Attention (MCMA), to improve prediction accuracy and latency.

Main Results:

• The MCMA model significantly outperformed the baseline mixed-effects model in predicting risk-taking behavior.
• MCMA achieved a Mean Absolute Error (MAE) of 3.17, Root Mean Squared Error (RMSE) of 4.38, and R-squared (R2) of 0.93, compared to the baseline's MAE of 10.97, RMSE of 14.73, and R2 of 0.30.
• Physiological activation and tactile interaction intensity were identified as prominent factors in risk processing during human-robot interaction.

Conclusions:

• Physiological data, combined with behavioral and tactile interaction data, can accurately predict human risk-taking behavior in human-robot interactions.
• The study demonstrates the feasibility of using machine learning for low-latency prediction of risk-taking behavior.
• Findings offer insights into the interplay between physiological responses, tactile interaction, and risk processing, informing the design of more intuitive and responsive social robots.