Automated derivation of behavior vocabularies for autonomous humanoid motion

In this paper we address the problem of automatically deriving vocabularies of motion modules from human motion data, taking advantage of the underlying spatio-temporal structure in motion. We approach this problem with a data-driven methodology for modularizing a motion stream (or time-series of human motion) into a vocabulary of parameterized primitive motion modules and a set of meta-level behaviors characterizing extended combinations of the primitives. Central to this methodology is the discovery of spatio-temporal structure in a motion stream. We estimate this structure by extending an existing nonlinear dimension reduction technique, Isomap, to handle motion data with spatial and temporal dependencies. The motion vocabularies derived by our methodology provide a substrate of autonomous behavior and can be used in a variety of applications. We demonstrate the utility of derived vocabularies for the application of synthesizing new humanoid motion that is structurally similar to the original demonstrated motion.