Crafting Realistic Virtual Humans: Unveiling Perspectives on Human Perception, Crowds, and Embodied Conversational Agents

TL;DR

This paper surveys how human perception shapes the realism of Virtual Humans (VHs) and their applications, focusing on the Uncanny Valley, computational measures of perceived discomfort, and biases in VH perception. It analyzes two key VH applications—crowd simulation and embodied conversational agents (ECAs)—reviewing modeling approaches from microscopic to macroscopic crowd dynamics and highlighting state-of-the-art ECAs built with tools like MetaHumans. The authors discuss ethical considerations and the need for inclusive design to mitigate stereotypes while outlining future directions in personalization and AI-driven interactivity. The work emphasizes practical implications for training, urban planning, and health-oriented interactions, where highly realistic VHs can enhance presence, engagement, and realism, provided biases and the uncanny valley are carefully managed.

Abstract

Virtual Humans (VHs) were first developed more than 50 years ago and have undergone significant advancements since then. In the past, creating and animating VHs was a complex task. However, contemporary commercial and freely available technology now empowers users, programmers, and designers to create and animate VHs with relative ease. These technologies have even reached a point where they can replicate the authentic characteristics and behaviors of real actors, resulting in VHs that are visually convincing and behaviorally lifelike. This paper explores three closely related research areas in the context of virtual humans and discusses the far-reaching implications of highly realistic characters within these domains.

Figures (3)

• Figure 1: All the characters used in the work of Flach et al. dill2012evaluation (left), and the perceptual comfort obtained with more recent characters (right). Both blue and orange lines on the left represent the percentages of the perceived comfort of each character in the image and video, as perceived in 2012. However, the green and yellow lines represent the same VHs evaluated in 2020. In addition, in (b), we can see the results regarding recent characters perceived in 2020 araujo2022perceived
• Figure 2: On the left, we observe a bottleneck scenario where the microscopic crowd is densely organized (people/$m^2$) as they pass through the door. In the center, we present an example of a macroscopic crowd simulation, demonstrating how the crowd avoids collisions with objects by dynamically splitting and merging before and after the collision event, respectively da2020towards. On the right, we glimpse a city designed to serve as the environment for a macroscopic population simulation silva2020lodus.
• Figure 3: Illustrations of facial and body animations designed for Arthur using Unreal and MetaHuman.