Oscillatory signals drive reaching and locomotion for both robots and primates. Recent results in neuroscience have shown that periodic signals are present in the motor cortex of primates during rhythmic tasks such as locomotion as well as during linear movements notably reaching. Our works on learning latent motion representations in robotics revealed that oscillatory latent dynamics emerge automatically from training data for quadruped locomotion as well as manipulation tasks. Inspired by these works, we recreate the locomotion latent-spaces as well as create manipulation specific versions. With the latter we show that manipulation problems can be solved using periodic signals in a suitable latent-space. We see that these trajectories are reminiscent of those seen in the motor cortex. We artificially lesion the decoder of the locomotion model to understand how the correlations are captured. This results in deformation of both the oscillatory signals in the latent space and degradation in the locomotion trajectories.