For motor control, the activities of local neural nets need to be coordinated, which can be attained via direct synaptic connections. Phylogenetically, the main purpose of the brain is the active interaction with the environment which suggests the presence of an alternative neural coordination mechanism, namely one that is mediated by a feedback loop involving the outside world. In this context we study the coordination of otherwise isolated neural clusters via 'force coupling', viz exclusively through the reactive forces of the body. Coordination arises when a given neural net reacts to the changes induced in the environment by the output of another net. Force coupling leads to self-organizing control principles, which we implement for simulated robots actuated by synthetic muscles. We show that stable limit cycles are generated when the embodied feedback loop is closed, giving rise in turn to robust regular locomotive gaits. As such, force coupling reduces the complexity of motor control tasks, defining at the same time a novel class of dynamical neural systems.