A survey on generative nonverbal facial behavior for highly realistic embodied agents

The rise of generative artificial intelligence has revitalized interest in its applications for realistic human–robot interaction. While verbal facial behavior generation is well developed and widely adopted in industry, its nonverbal counterpart remains underexplored and poorly integrated into robotics, despite its crucial role in seamless communication. Research in this area is fragmented, and most state-of-the-art androids and virtual avatars still rely on outdated nonverbal facial behavior models. This paper reviews the current state of the hyperrealistic embodied agents and their endeavors in nonverbal facial behavior research, identifying key challenges that hinder its broader adoption and development. We categorize and analyze critical factors influencing model design, including facial representation, generative modeling, and existing android/virtual avatar platforms. Our goal is to provide new researchers with a curated overview of state-of-the-art techniques and potential platforms for feature enhancement. By offering systematic guidelines and comparative analysis across different branches of generative nonverbal facial behavior, we aim to bridge the gap between research and application, improving nonverbal human–robot communication in both androids and virtual avatars.