CS Events
PhD DefenseContext-Sensitive Narrative Generation for Virtual Populations and Application to Human-Building Interaction |
|
||
Wednesday, June 14, 2023, 10:00am - 12:00pm |
|||
Human crowd behavior simulation is an indispensable tool in a myriad of sectors including, but not limited to, film production, urban design, and virtual evacuation simulation. It serves as a powerful mechanism for both production and testing. The primary objective of this dissertation is to address issues pertaining to the narrative generation of virtual crowds in the context of semantically rich, data-driven environments and enhance computer-aided design workflows through the utilization of information amassed during simulation.
Although prior studies have made strides towards enhancing visual fidelity and runtime efficiency of crowd behavior simulation, only a handful have endeavored to simulate the behaviors of individual agents using real-world semantic information as a constraint to augment the authenticity of the simulation. This dissertation redefines crowd behavior simulation as a problem of environmental scale, viewing each agent's behavior as an integral component of a larger narrative. This necessitates meticulous behavior management, ensuring all agents are environment-aware and consistently exhibit plausible behaviors, compliant with the pertinent semantic information within the scene.
This dissertation develops innovative frameworks for creating context-aware virtual agents in interactive narratives. Starting with introducing an accelerated partial order planner for real-time narrative generation and repair that adapts to user interactions with minimal re-planning, a major contribution of this work is a memory reconstruction method for virtual agents that captures and transforms agents' memories, behaviors, and interactions into coherent narratives. This process enhances the immersion and depth of the narrative experience. A comprehensive framework is proposed for authoring behavior narratives, considering spatial, behavioral, and agent-centric features. This approach dynamically models behavior narratives while accommodating multi-level constraints, thereby improving the realism of agent behavior. We employ an ontology graph derived from a commonsense knowledge corpus to infer behavior distributions, enabling user-interactive refinement of a virtual environment's design. A vital strategy is proposed to minimize airborne disease exposure, such as COVID-19, in indoor environments using a novel social distancing efficiency metric. This optimization strategy has practical implications for controlling disease diffusion.
This work contributes to the advancement of virtual agent research, with potential impacts on interactive storytelling, gaming, architectural design, and public health.
Speaker: Xun Zhang
Location : Virtual
Committee:
Professor Mubbair Kapadia (Chair)
Professor Mridul Aanjaneya
Professor Jingjin Yu
Professor Nuria Pelechano (Polytechnic University of Catalonia)
Event Type: PhD Defense
Abstract: See above
Organization:
Rutgers University
School of Arts & Sciences
Department of Computer Science
Contact Professor Mubbasir Kapadia