Abstract
A common task in dance, martial arts, animation, and many other movement genres is for the character to move in an innovative and yet stylistically consonant fashion. This paper describes a corpus-based method for automatically generating such movement sequences. Our algorithms use the mathematics of chaos to achieve innovation and simple machine-learning techniques to enforce stylistic consonance. We first construct a mapping between an original motion sequence and a chaotic attractor. Chaos's hallmark sensitive dependence on initial conditions guarantees that variations generated from this mapping are different from the original; the fixed structure of the attractor guarantees that the variations resemble the original in both esthetic and mathematical senses. The effects of this mimic a choreographic variation technique that dates from the early 1960s: pasting together segments of a given dance in a new order. As one would imagine, that mechanism can create abrupt transitions at the segment boundary, which calls for some smoothing. Simple interpolation-cf., morphing-is not a good solution here, as it is guided by mathematical constraints (smoothness, shortest path, etc.) and so its results are not necessarily consistent with a specific movement style. Our algorithms employ simple graph-theoretic methods to learn the `grammar' of joint movements in a given corpus, and then search those graphs to find stylistically consistent interpolation sequences between pairs of body postures.
Full paper here.
Also available as Department of Computer Science Technical Report CU-CS-1029-07..