Combining Motion Capture Data with PD Controllers For Human-like Animations

Motion capture data has rich details and styles but it is difficult to modify and reuse. And keyframing provides much more freedom to designers but requires the high level of skills and efforts. On the other hand, dynamics simulation has shown impressive results of rigid body simulation in games and movies. In this paper, we propose a new method to transform motion capture data effectively using dynamics simulation and to synthesize natural and realistic motions. Because people can separate virtual characters from real actors with a few frames of motion, our approach employed motion capture data as a reference motion. Also in order to provide users controllability of behaviors of virtual characters, we applied two types of controllers; local and global controllers. A local controller attached to every joint produces torque for motion tracking, while a global controller attached to the root joint generates torque and force for the overall root trajectory. With these two types of controllers, meeting the final pose and connecting to another motion capture data is also possible. Our approach is simple rather than complex optimization and delicate modeling of controllers. But experimental results show the motions of virtual characters are natural and realistic