CPA-G&V: Physically Valid Simulation of Active Human Motion

（Jovan Popovic，麻省理工学院）交互式计算机图形模拟为学习复杂的关系和过程提供了一种行之有效的方法。例如，飞机模拟器指导飞行员完成复杂的任务，直到所有程序都安全有效地执行。对这些应用来说，特别重要的是与现实世界一致的模拟。例如，在医学领域，医生可以掌握&amp；#64257；在不危及患者生命的情况下，在生理上一致的模拟中提高运动技能。然而，目前使用的模拟只能模拟反应性物理系统。计算机模拟在医学、工程、艺术和娱乐领域的广泛应用需要模拟主动系统，如机器人设备、人类和动物，这些系统在物理模拟的环境中起作用和反应。例如，一项针对车辆和工作空间设计的模拟研究表明，人们普遍需要能够在物理模拟环境中完成任务的数字人体模型。今天的计算机图形学经常把人类模拟成没有生命的机械结构（布娃娃），只不过是在地上皱巴巴的。或者它依赖于动作捕捉技术（动作图），每当模拟偏离一些动作捕捉场景时，就会产生物理上不一致的动作。相比之下，基于物理的模拟产生一致的动作，但这些动作通常是不自然的，很难创造。本研究运用现代控制理论的数学，为交互式计算机模拟中此类运动的控制建立更严格的基础。它通过稳定高质量的动作捕捉数据来模拟自然动作，将单个动作转化为可应用于许多不同情况的更一般的动作。单个动作可以描述短任务，如踏步、转弯、行走和跳跃。更复杂的行为是通过组合，通过一个动作接一个动作来组合的。

Abstract (Jovan Popovic, MIT)Interactive computer-graphics simulations offer a proven method for learning complex relationships and procedures. Airplane simulators, for example, tutor pilots through sophisticated tasks until all procedures are performed safely and efficiently. Of particular importance to these applications are simulations that are consistent with the real world. In medicine, for example, physicians can master fine motor skills within physiologically consistent simulations without endangering the lives of patients. However, the simulations in use today can only emulate reactive physical systems. Broader application of computer simulation in medicine, engineering, art, and entertainment requires simulation of active systems such as robotic devices, humans, and animals that act as well as react within physically simulated environments. A study of simulations for vehicle and workspace design, for example, has revealed a pervasive need for digital human models that can accomplish tasks within physically simulated environments. Computer graphics today often simulates humans as lifeless mechanical structures (rag dolls) that do no more than crumple to the ground. Or it relies on motion-capture techniques (motion graphs) that generate physically inconsistent motions whenever simulations depart from a few motion-captured scenarios. In contrast, physically based simulation generates consistent motions, but the motions are often unnatural and difficult to create. This study applies the mathematics of modern control theory to establish a more rigorous foundation for control of such motions in interactive computer simulations. It simulates natural motions by stabilizing high-quality motion-capture data, transforming individual motion into a more general action that can be applied in many different circumstances. A single action may describe short tasks such as stepping, turning, walking, and jumping. More complex behaviors are assembled through composition, by concatenating one action after another.