Introducing Agent Personality in Crowd Simulation Improves Social Presence and Experienced Realism in Immersive VR

Convincing crowd behavior simulation is becoming essential in many application domains, including video games, cinematography, urban planning, safety simulations, and training. In this paper, we propose a novel and lightweight mesoscopic system for personality-based crowd simulation in immersive virtual reality (iVR). We use the Big Five personality framework, also known as OCEAN, to model a synthetic personality for each autonomous agent. Agents can autonomously aggregate in formations using machine learning-based clustering techniques operating on OCEAN. Moreover, agents can also externalize their personality traits by performing peculiar behavioral animations. To choose which animations to perform, we adopt a probabilistic approach that considers each OCEAN dimension as a continuous spectrum with two extremes linked to pairs of animations. Our system is designed to be flexible and suitable for different applications. Flexibility is achieved by using graphs to store agent and map topology data that control how the agents move and behave at runtime. In a within-subjects study with 40 users, we compare our personality-based system against a basic system that does not use personality. Results show that introducing personality into iVR crowd simulation enhances users' social presence and experienced realism. Introducing personality also increases the perceived match between the agents and the virtual environment where the simulation takes place.