CGV: Small: Simulation Motion Capture of Dexterous Manipulation

这项研究涉及到捕捉自然人类操纵行为的所有复杂程度的新技术。想想我们如何准备食物、拿起和使用扳手之类的常用工具，或者种一朵花。我们希望向人类学习这种复杂动作的例子，这样我们就可以描绘出令人信服的人类角色的图形再现，应用范围从教育、培训和娱乐到博览会和科学研究。更广泛的影响包括使用数据库和作为这项研究的一部分开发的捕获工具，用于开发灵活的机器人操作、假肢设计以及对人类抓取和操作的理解。这项研究的视觉性质和吸引力使其特别适合于指导本科生和女性，以及推广到激发K-12学生对科学的兴趣。传统和既定的运动捕捉方法，如光学方法，无法处理必须遵守的复杂、详细、高度自由的交互来捕捉灵活的操作动作。这项研究中追求的另一种选择是模拟运动捕捉，用户与运行中的模拟交互并引导模拟。使能技术包括(1)与模拟对象密切交互的骨骼驱动的可变形手的实时模拟，(2)允许直接控制这种模拟的新的控制算法和接口技术，(3)基于模拟期间的用户动作的用于分段和控制律开发的新语言，以及(4)从直接观察人类动作学习的新的操纵目标函数。与当前的捕获技术相比，模拟捕获的优势包括很少或没有后处理、接触力的精确表示、快速试验和错误、能够利用各种方法进行控制以及能够使用各种输入设备。由此产生的捕获系统将是廉价的，而且可以随时使用。

This research involves novel techniques for capturing natural human manipulation actions in all of their complexity. Consider how we prepare food, pick up and use a common tool such as a wrench, or plant a flower. We would like to learn from human examples of such complex actions, so that we can portray convincing graphical renditions of human characters for applications ranging from education, training, and entertainment to exposition and scientific study. Broader impact includes use of databases and the capture tool developed as part of this research for development of dexterous robotic manipulation, design of prosthetics, and understanding of human grasping and manipulation. The visual nature and appeal of this research makes it especially suited for mentoring of undergraduates and women, as well as for outreach to spark interest in science for K-12 students.Traditional and established approaches to motion capture, such as optical approaches, cannot handle the complex, detailed, high degree of freedom interactions that must be observed to capture dexterous manipulation actions. The alternative pursued in this research is simulation motion capture, where a user interacts with and guides a running simulation. Enabling technologies include (1) real-time simulation of a skeleton driven deformable hand in close interaction with simulated objects, (2) novel control algorithms and interface techniques to allow direct control of such simulations, (3) a new language for segmentation and control law development, based on user actions during the simulation, and (4) novel manipulation objective functions learned from direct observation of human actions. Benefits of simulation capture over current capture technologies include little or no postprocessing, precise representation of contact forces, fast trial and error, ability to make use of a variety of means for control, and ability to use a variety of input devices. The resulting capture system will be inexpensive and broady accessible.