My work enables users to create complex and realistic animation by providing a small amount of information. For example, a human animation might be specified by posing an articulated doll or sketching a path through a complex environment. The two algorithms I have developed can automatically create natural-looking human motion that matches this specification.  The key insight behind both approaches is to build a compact representation of human motion based on available human motion capture data. The first approach builds a continuous low-dimensional representation, while the second builds a discrete representation consisting of only natural poses. The compactness of these representations allows an optimizer to efficiently find a solution that matches the user's specification. In addition, these representations favor natural poses and velocities, which biases the optimizer towards natural-looking solutions---an objective that is very hard to define mathematically. In my talk, I will compare and demonstrate the effectiveness of these two approaches in synthesizing natural, physically realistic, complex motions sketched using a simple interface.

Tuesday, March 27, 2007

3:00 pm - 4:15 pm

Wu & Chen Auditorium

101 Levine Hall