AUTONOMOUS SOCIAL ROBOTS: LEVERAGING ARTIFICIAL INTELLIGENCE FOR INDEPENDENT DECISION-MAKING IN COMPLEX HUMAN-ROBOT INTERACTION SCENARIOS

Abstract

This study explores how artificial intelligence decision-making models shape the performance and social acceptance of Autonomous Social Robots (ASRs). Focusing on Reinforcement Learning (RL), Deep Learning (DL), and Rule-Based (RB) approaches, we employed a within-subjects experimental design in which 90 participants interacted with robots across simple, mixed, and complex tasks. Quantitative data were gathered on task completion time, decision accuracy, trust, and emotional comfort, while qualitative observations captured subtle user reactions. Results demonstrate that RL-enabled robots consistently outperformed DL and RB models in efficiency, adaptability, and decision accuracy. RL systems were also perceived as more trustworthy and socially competent, particularly in dynamic scenarios. By contrast, RB models performed reliably only in structured environments but failed to accommodate emotional and contextual nuances, resulting in reduced trust. DL showed moderate adaptability but was limited by perceived opacity in decision-making. Findings confirm that user trust and emotional comfort are tightly linked to robot adaptability, underscoring the necessity of socially intelligent, ethically transparent AI systems. This research contributes to both theory and practice by providing evidence for the superiority of RL in socially complex settings and by highlighting the ethical and design imperatives needed to foster trust-based human-robot rapport in real-world applications.