CAREER: Synthesis of Autonomous, Realistic Human Motion

在虚拟环境中对可信的人类角色进行建模，对于释放计算机作为交互媒介的潜力至关重要。对人体运动的逼真描述极大地增强了任务训练、社会交流或远程学习的知觉沉浸感。在可信的虚拟环境中，人类角色必须能够规划自己的运动，对环境做出反应，并以逼真的情感和个性展示吸引其他虚拟角色或真实的人类。这项艰巨的任务需要从肌肉骨骼系统到神经系统的模拟，因为当进行功能活动时，这些复杂的组成部分必须同步协调。今天，基于物理学的角色动画可以模拟复杂的运动序列，具有令人惊叹的视觉真实感，尽管是在动画师的精心指导下。大多数角色动画技术缺少的是自主控制，无论是有意识的还是无意识的。在这个项目中，研究人员将计算机生成的角色动画从一个可视化工具扩展到一个专注于自主控制和真实运动的跨学科研究领域。本项目关注两个相互关联的研究推动力，从根本上与研究人员的研究主题的跨学科性质相一致：(1)合成无弹道、低能量的人类运动，以及(2)合成多个角色之间的互动。与弹道和高动态运动不同，非弹道运动不是由物理严格确定的，因此需要更复杂的模型来处理无意识自主控制(如生理和情感)和有意识自主控制(如角色意图)的复杂相互作用。当多角色互动开始发挥作用时，自主控制变得更加关键。交互的合成是具有挑战性的，因为当交互发生时，高级人工智能算法和低级移动以不可预测的方式交织在一起。在计算机动画的范围之外，这个计算模型可以提供一种识别情绪、诊断肌肉骨骼异常或验证生物力学假说的设备。

Modeling believable human characters in the virtual environment is crucial for unleashing the potential of computers as an interactive medium. A realistic depiction of human motion drastically enhances the perceptual immersion for task training, social communication, or tele-learning. In a believable virtual environment, a human character must be able to plan its locomotion, react to the environment, and engage other virtual characters or real humans with realistic display of emotion and personality. This daunting task requires simulation from the musculoskeletal system to the nervous system, as these intricate components must coordinate in synchrony when a functional activity is undertaken. Physics-based character animation today can simulate complex locomotion sequences with stunning visual realism, albeit under the careful guidance of the animator. What is missing from most character animation techniques is autonomous control, both conscious and unconscious. In this project, the investigator expands computer-generated character animation from a visualization tool to an interdisciplinary research area focused on autonomous control and realistic locomotion.This project focuses on two inter-related research thrusts fundamentally aligned with the interdisciplinary nature of the investigator's research theme: (1) synthesis of non-ballistic, low-energy human locomotion, and (2) synthesis of interaction among multiple characters. Unlike ballistic and highly dynamic motion, non-ballistic motion is not stringently determined by physics, thus requiring more sophisticated models to address the complex interplay of unconscious autonomous control, such as physiology and emotion, and conscious autonomous control, such as the intents of the character. The autonomous control becomes even more crucial when multiple character interaction comes into play. Synthesis of interaction is challenging because the high-level artificial intelligence algorithms and low-level locomotion intertwine in an unpredictable manner when the interaction takes place. Beyond the scope of computer animation, this computational model can provide a device for recognizing emotion, diagnosing musculoskeletal anomalies, or validating biomechanical hypotheses.