Towards a Live Dance Improvisation between an Avatar and a Human Dancer

This paper presents an attempt to generate novel dance movement based on motion captured human dance. Captured movements are analyzed statistically using nonlinear principal component analysis in order to create a "map" of observed poses. A method for automatically exploring the map by generating random trajectories is then presented, constituting a kind of improvisation. The result is an animated avatar that exhibits creative and novel movements in the style of its teacher, and paves the way towards a fully improvised live performance between an avatar and a human dancer.