Optimizing Virtual Reality Multi-Character Experiences Using Affective Ratings

Angshuman Mazumdar, Purdue University

Abstract

The thesis deals with the study of how virtual multi-character scenarios (primarily crowds) can be synthesized with specific behaviors, so as to induce a negative affect in the user. Virtual crowds are inclined towards being a passive world building factor, rather than a gameplay affecting factor. The study focuses on one main research question: “Is it possible to synthesize a multi-character experience that induce a certain amount of negative affect to participants?” Through the study, the emphasis lies on being able to drive emotions in an effective way, when creating multi-character scenes that need to give off a specific mood or emotion and provide an insight into how the behavior of the collective is able to affect a user’s mindset. The pipeline’s development involved developing a dataset of behaviors to be assigned to the virtual characters. Next an annotation phase assigned the affective scores to the virtual behaviors (34 in total), which (along with several design parameters) were then considered for the total cost of a scenario with a multi-character setup. Using a Markov chain Monte Carlo technique known as Simulated Annealing, the scenes were optimized towards target values of negative affect (namely low, medium, and high target affects). Finally, through the implementation of a user study, the algorithm was validated on synthesizing these targeted affect-driven multi-character virtual reality experiences. The results indicated that the three synthesized experiences (low, medium, and high negative affects) were perceived as expected by participants. Thus, the study concluded by stating that affect-driven multi-character virtual reality experiences can be automatically synthesized in such a way that impacts a user’s affect levels in the way that is expected.

Degree

M.Sc.

Advisors

Adamo, Purdue University.

Subject Area

Information Technology

COinS