Humans can learn bimodal priors in complex sensorimotor behaviour

Bayesian integration offers a powerful unifying framework for perception, cognition and motor control. Extensive laboratory-based research shows that humans integrate sensory inputs and prior expectations in a Bayesian manner to guide action. However, it remains unclear whether this extends to naturalistic tasks involving more complex movements and environmental statistics. Here, we examine whether humans can learn bimodal priors in a complex sensorimotor task: returning tennis serves. Participants returned serves in an extended reality set-up with task demands closely matching real tennis. The opponent's serve locations followed a bimodal (i.e. two-peaked) distribution. We manipulated visual uncertainty through three ball speeds. After extensive exposure to the opponent's serves, participants' movements were systematically biased by the bimodal prior distribution. Consistent with Bayesian theory, the magnitude of the bias scaled with uncertainty. In addition to prior expectations, our data show that participants' movements were also biased by biomechanical constraints and motor costs. Intriguingly, explicit knowledge tests revealed that participants incorporated prior knowledge of the opponent's serve distribution into their behaviour without explicit awareness of the pattern. These findings demonstrate that humans can implicitly acquire and exploit bimodal priors in complex, naturalistic sensorimotor behaviour.