During joint actions, humans continuously exchange coordination signals and use nonverbal, sensorimotor forms of communication. Here we discuss a specific example of sensorimotor communication-"signaling"-which consists in the intentional modification of one's own action plan (e.g., a plan for reaching a glass of wine) to make it more predictable or discriminable from alternative action plans that are contextually plausible (e.g., a plan for reaching another glass on the same table). We first review the existing evidence on signaling in human-human interactions, discussing under which conditions humans use signaling. Successively, we distill these insights into a computational theory of signaling during online interactions. Central to our approach are the following ideas: 1) signaling endows pragmatic plans with communicative goals; 2) signaling can be understood within a cost-benefit scheme, balancing the costs for the signaling agent against its benefits for interaction success; and 3) signaling may be part of an interactive strategy that optimizes success when joint goals are uncertain. Finally, we exemplify the benefits of signaling in a series of simulations and discuss how endowing robots with signaling abilities can increase the quality of human-robot interactions by making their behavior more predictable and "legible" for humans.
