The kinematic Motion Primitives (kMPs) of periodic motions, discrete motions, and motions that are a combination of discrete and periodic movements

The research presented in this manuscript aimed to: i) analyze human motion, introducing the kinematic Motion Primitives (kMPs), ii) propose an application of the kMPs to synthesize human-like walking trajectories for the COmpliant huMANoid COMAN. A Principal Component Analysis (PCA) was applied to the whole-body joint trajectories, that were derived from human data recorded with a motion capture system. Five kMPs were identified for the periodic motions (locomotion), that are invariant to the subjects, to the walking/running velocity, and to the constraints imposed on the swing motion of the arms. Two kMPs are sufficient to describe the discrete motions (reaching for a target with one hand), and are invariant to the subject, and to the hand used for reaching. These two sets of kMPs can combine, and define those motions that are a simultaneous combination of periodic and discrete movements (reaching while walking). As a demonstration of possible application, human-like trajectories for walking were generated by reconstruction from kMPs, and successfully tested on the COMAN robot