Diverse Synthetic Crowds in Media, Design, and Analysis

我的研究计划解决了越来越多的代表性，公平性和包容性的合成人群的需求。合成群体是模拟许多人的运动，行为和互动的实践。人类的模拟对于许多研究和应用活动都是极其重要的。这包括从电影和电视角色的大规模动画到安全关键场景的测试和重建的一切。过去实现这一目标的方法主要集中在使用人类的简化表示来创建规范人群的引人注目的自然主义运动。这些模型已被应用于工具链中，这些工具链不能充分满足评估、模型的上下文感知选择以及包含不同模拟人类的需求。这意味着合成人群在很大程度上不代表个人，如果他们不具有代表性，那么使用这些合成人群的设计，媒体和其他应用程序就不具有代表性。在拟议的计划中，我专注于增加人类及其活动的保真度和范围，在人群动画，环境设计，以及环境和政策分析的方法。 所提出的研究的目标是从根本上改变模型，工具集和实践中的合成人群的研究和应用。拟议的方法有三个方面。(1)从具有不同行动能力和行动辅助设备的人那里收集有代表性的数据，并为公众访问提供丰富的元级信息。这包括开发运动捕捉方法和丰富的个性化数据收集协议，包括各种运动动作，移动性和移动设备。(2)在模拟中更广泛、更多样化地考虑人类的流动性。这包括采用和开发机器学习中的新方法，这些方法既受数据驱动又受模拟驱动，以学习高保真物理模拟人类的个体、步态和辅助设备的不同表示，同时学习数据未捕获的人群交互。(3)增强和重新设计工业工具链。这包括开发新的设计和分析方法，使从业者和政策制定者能够探索其设计的可访问性和公平性。这项工作的核心是为工业和社区开发工具，以便通过包容性做法，如参与性和/或社区驱动的设计和分析，调查和设计更好的环境。这还包括游戏的开发和增强的设计和分析方法。这些方法有望快速扩展工程能力，同时增加历史上非包容性设计管道的包容性。

My research program addresses the increasing need for representation, equity, and inclusion in synthetic crowds. Synthetic crowds are the practice of simulating the movement, behaviours, and interactions of many humans. The simulation of humans is extremely important to several research and applied activities. This includes everything from the large scale animation of characters for film and television to the testing and recreation of safety-critical scenarios. Past approaches for achieving this are largely focused on creating compelling and naturalistic movements of normative crowds using simplified representations of humans. These models have been applied in toolchains which do not adequately address the need for evaluation, context-aware selection of models, and the inclusion of diverse simulated humans. This means that synthetic crowds are largely not representative of individuals, and if they are not representative, then designs, media, and other applications using these synthetic crowds are not representative. In the proposed program, I focus on increasing the fidelity and scope of humans and their activities in methods for crowd animation, environment design, and both environment and policy analysis. The goal of the proposed research is to radically transform the models, toolsets, and practices in synthetic crowd research and applications. The proposed methodology is three-fold. (1) The collection of a representative amount of data from people with diverse mobilities and mobility assistive devices, with rich meta-level information purpose-built for public access. This includes developing motion capture methods and rich individualistic data collection protocols that include a variety of locomotion actions, mobilities, and mobility devices. (2) A broader more diverse consideration of human mobility in simulation. This includes adopting and developing new methods in machine learning that are both data- and simulation-driven to learn diverse representations of individuals, gaits, and assistive devices for high-fidelity physically simulated humans, while simultaneously learning crowd interactions not captured by the data. (3) The augmentation and redesign of industry tool chains. This includes developing new methods in design and analytics which enable practitioners and policymakers to explore the accessibility and equity of their designs. Central to this effort is the development of tools for both industry and communities to investigate and design better environments through inclusive practices, such as participatory and/or community-driven design and analysis. This also includes the development of games and augmented approaches to design and analysis. These approaches promise to rapidly expand engineering capabilities while increasing inclusivity in a historically non-inclusive design pipelines.