In this paper, we present a novel approach to dynamically describe human upper limb trajectories, addressing the question on whether and to which extent synergistic multi-joint behavior is observed and preserved over time evolution and across subjects. To this goal, we performed experiments to collect human upper limb joint angle trajectories and organized them in a dataset of daily living tasks. We then characterized the upper limb poses at each time frame through a technique that we named repeated-principal component analysis (R-PCA). We found that, although there is no strong evidence on the predominance of one principal component (PC) over the others, the subspace identified by the first three PCs takes into account most of the motion variability. We evaluated the stability of these results over time, showing that during the reaching phase, there is a strong consistency of these findings across participants. In other words, our results suggest that there is a time-invariant low-dimensional approximation of upper limb kinematics, which can be used to define a suitable reduced dimensionality control space for upper limb robotic devices in motion phases.