X's Day: Personality-Driven Virtual Human Behavior Generation

This IEEE conference talk presents a novel framework for generating long-term behaviors for virtual agents based on personality traits and environmental context. Learn about the hierarchical structure connecting personality characteristics to observable activities through Needs, Task, and Activity levels. The presentation demonstrates how the integration of a Behavior Planner and World State module enables dynamic behavior sampling using large language models (LLMs), allowing virtual humans to respond appropriately to environmental changes. The research focuses on autonomous production of daily activities in a 3D environment, creating more realistic and contextually appropriate virtual human behaviors. Visit the project website at https://behavior.agent-x.cn/ for more information about this innovative approach to personality-driven virtual human behavior generation.