Exploiting prior knowledge in continuous decision-making under uncertainty: the case of tennis experts.

Leading theories of sensorimotor control propose that humans navigate uncertainty by integrating prior and sensory information in a Bayesian manner. However, empirical evidence is largely limited to static and constrained lab tasks. How humans exploit prior knowledge during continuously unfolding decision-making in naturalistic behavior remains unclear. Here, we study a task that pushes the human sensorimotor system to its limits: returning fast tennis serves. This task is particularly informative for gaining insights into the dynamics of unfolding decisions in action, owing to a distinct feature of the return movement: the split step—a preparatory movement characterized by a small jump to increase initial speed in the desired direction. In the experiment, experienced tennis players returned serves in an immersive extended reality setup with unconstrained movements and task demands matching real tennis. We manipulated the distributions of the opponent’s preferred serve locations (80% to the right vs. 20% to the left of the service box and vice versa in a second session). As a continuous behavioral readout of the evolving decision-making process, we measured participants’ weight shift during the unfolding action. Results show that, over the experiment, participants increasingly rely on acquired prior knowledge of the more probable serve direction to improve performance. Participants exploit prior knowledge by shifting their weight toward the more probable serve direction—already before the serve—while continuously re-evaluating both options based on incoming sensory information and motor costs. Using tennis as an exemplary case, our findings provide evidence that humans use prior and sensory information to probabilistically optimize continuous decision-making in complex sensorimotor behavior.