When: Monday, May 05, 1:00 p.m.

Where: 5409Wean Hall

Manfred Lau

Thesis Proposal

Abstract:
The simulation of crowds of virtual characters is needed for applications such as films, games, and virtual reality environments. These simulations are difficult due to the large number of characters to be simulated and the requirement for synthesizing realistic human-like motions efficiently. We focus on two problems: how to model motion clips of behaviors so that we can generate human-like motions for multiple characters interactively, and how to model and synthesize variations in motion data.

To generate motions for multiple characters navigating autonomously in large and dynamic environments, we model motion clips as high-level behaviors and develop a single global planning scheme that applies a collection of motion planning techniques. Moreover, we explore the idea of precomputing a set of motion paths to speed up the motion synthesis process. This precomputation allows us to generate motions much faster than before, and enables us to develop an interactive system with a large number of characters. Within this framework, we propose to explore further the important issue of how to build a smaller and diversified tree of motion paths.

Current state-of-the-art crowd simulations often use specific motion clips that are repeatedly replayed or a few cycles of a particular motion to continuously animate multiple characters. The idea of synthesizing the subtle variations in motion data has not been explored. We would like to argue that variation is not just an additive noise component, as presented in previous approaches. Instead we learn a generative model of motions from example data, and use this model to synthesize new variations of the input data. We use a Dynamic Bayesian Network model, and we explain why this model is particularly suited for solving our problem. Our preliminary results have shown that we can synthesize new variations that are natural and statistically valid. We identify a number of problems to be solved within our current framework.

Thesis Committee:
James Kuffner, Chair
Nancy Pollard
Ziv Bar-Joseph
Ming Lin, University of North Carolina at Chapel Hill

Thesis Summary: http://www.cs.cmu.edu/~mlau/mlau_thesis_proposal.pdf

<< Back