Efficient synthesis of physically valid human motion for computer animation.

摘要

Animation is a powerful storytelling medium, within which artists can develop themes relating to the viewer's own experiences and physical existence. The plausibility of an animation depends on both objective criteria, such as obeying the laws of physics, and subjective criteria, such as exhibiting a consistent style of motion. One appealing vision for animation systems is that they somehow take care of the objective criteria, while allowing artists as much freedom and expressive power as possible.; One approach toward this vision has been to compute a motion which optimizes energy expenditure while meeting both physical laws and animator constraints. Optimal energy expenditure has some nice properties, such as providing the appearance of squash-and-stretch in the jumping motion of a simple character, but extending these results to humanlike characters has proven difficult—the optimization process is often viewed as being slow, scaling poorly with character complexity, and being difficult to control.; Much may be gained by focusing on physical validity rather than the optimality of physically based quantities. This thesis shows that an approach focusing on physical validity rather than physical optimality can improve the theoretical complexity of each iteration of a numerical optimization routine from O(N2) to O( N), where N is the number of degrees of freedom of the character. Because there is a larger space of physically valid solutions than physically optimal solutions, this approach in principle also leaves more room for animator control.; The animation system is particularly useful for synthesizing highly dynamic motions. Results are shown of humanlike characters having from 7 to 22 degrees of freedom performing a variety of motions, including swinging, leaping, and back-flipping off a high-bar.