Motion segmentation for humanoid control planning

The discovery of major management behaviours from human motion data and uncovering their underlying components is investigated. A range of methods for segmenting major shifts in multidimensional time series are compared in inducing plausible behaviours from motion data. These behaviours are considered as supersets of motion primitives that define a repertoire of manoeuvres available to the human. The resulting multilayered symbolic model is used as a framework for humanoid imitation and control. It is hoped that with appropriate matching and scaling of degrees of freedom, models can be tested by extracting a trajectory for a simulation of the Nao soccer bot.