Controlling the body – given its huge number of degrees of freedom – poses severe computational challenges. In computational modeling and robotics, it is widely assumed that a control scheme using synergies simplifies movement planning and execution. Using synergies yields two useful forms of abstraction compared to the control of individual muscles or plant states: spatial abstraction, because a synergy specifies only some relevant aspects of the state of the plant (e.g., only the fingertip positions over time, not all the joint angles), and temporal abstraction, because a synergy corresponds to a period of time. We discuss why these two forms of abstraction are useful in action observation domains (not only motor control), considering computational architectures of action recognition and MNs that use a common grounding bayesian generative scheme.