While everyday actions flexibly combine rhythmic and discrete movement elements, motor neuroscience research has largely studied these movement types in isolation, developing separate accounts for these two movement forms. Supported by a series of brain-imaging, behavioral and modeling studies, our premise is that rhythmic and discrete behaviors are two dynamic primitives constituting more complex actions. Our research analyzes how task dynamics constrain and enable actions and their improvement with practice. We start with a mechanical model of the task and render it in a virtual environment with a fully known solution space. Based on mathematical analyses of the modeled task, we study how humans develop solutions to meet complex task demands. Key concepts in our analysis are variability, stability, and predictability. Using three model tasks, throwing a ball, rhythmic bouncing of a ball, and transporting a cup of coffee, we show that humans develop skill by: 1) finding error-tolerant strategies and channeling noise into task-irrelevant dimensions, 2) exploiting solutions with dynamic stability, 3) optimizing predictability of object dynamics. These findings are the basis for developing propositions about the controller: complex actions are generated with dynamic primitives, modules that overcome substantial delays and noise in the neuro-mechanical system. Extending from these experimental platforms we have developed interventions that assess or help restore functional behavior in neurological patients.