Author ORCID Identifier

0000-0001-7965-9921

Document Type

Dissertation

Date of Award

5-31-2026

Degree Name

Doctor of Philosophy in Information Systems - (Ph.D.)

Department

Informatics

First Advisor

Tomer Weiss

Second Advisor

Mark Cartwright

Third Advisor

Salam Daher

Fourth Advisor

Taro Narahara

Fifth Advisor

Yi Guo

Abstract

Simulating realistic crowd motion remains a fundamental challenge in computer graphics and multi-agent systems, as it requires modeling both physically plausible interactions and perceptually natural behaviors. Existing crowd simulation methods typically employ simplified geometric abstractions, most commonly circular agent representations, and model navigation using either analytical interaction formulations (e.g., force, velocity, or constraint-based methods) or learned policies derived through reinforcement learning. Despite their effectiveness, these approaches often overlook detailed geometric structure and do not explicitly account for perceptual realism. This dissertation addresses these challenges by improving the realism of virtual crowd simulation through two key advancements: perceptual preference learning and geometry-aware physical modeling.

The first contribution introduces a reinforcement learning—based framework that learns navigation policies optimized for perceptual realism. Instead of relying solely on manually defined reward functions, this approach incorporates human feedback through a crowd-sourced Bayesian preference learning mechanism to identify policy parameters that produce visually natural crowd behaviors. By integrating preference inference with deep reinforcement learning, the framework enables agents to exhibit anticipatory collision avoidance, smooth motion, and natural spacing, while aligning simulation outcomes with human perceptual judgments. This formulation provides a principled mechanism for learning navigation behaviors that are both physically plausible and perceptually preferred.

The second contribution addresses limitations in conventional geometric repre-sentations of crowd agents. Most existing methods approximate pedestrians as discs, which neglect orientation and overestimate spatial occupancy, limiting the fidelity of local interactions. This dissertation introduces a geometry-aware simulation model that represents agents as oriented capsules and formulates interaction dynamics using position-based constraints. The proposed constraint formulations capture translational and rotational avoidance behaviors, enabling agents to realistically adjust their orientation, shoulder alignment, and spatial footprint during navigation. This representation significantly improves the accuracy of interaction modeling, particularly in dense and constrained environments, while maintaining computational efficiency suitable for interactive simulation.

Overall, this dissertation contributes two distinct advances toward realistic virtual crowds. The first introduces a preference-driven learning framework that incorporates crowd-sourced perceptual judgments into reinforcement learning for navigation policy selection. The second develops a geometry-aware, constraint-based simulation model using oriented capsule agents to capture orientation-dependent, physically plausible local interactions. Although these contributions are developed independently, together they broaden the methodological foundations for crowd simulation by addressing realism from both perceptual and geometric perspectives, with applicability to computer graphics, interactive virtual environments, and related multi-agent simulation domains.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.