Synthesizing physically realistic human motion in low-dimensional, behavior-specific spaces

摘要: Optimization is an appealing way to compute the motion of animated character because it allows user specify desired in a sparse, intuitive way. The difficulty solving this problem for complex characters such as humans due part high dimensionality search space. artifact representation most dynamic human behaviors are intrinsically low dimensional with, example, legs and arms operating coordinated We describe method that exploits observation create optimization easier solve. Our utilizes existing capture database find low-dimensional space captures properties behavior. show when solved within subspace, sparse sketch can be used initial guess full physics constraints enabled. demonstrate power our approach with examples forward, vertical, turning jumps; running walking; several acrobatic flips.