Humans can learn bimodal priors in complex sensorimotor behaviour

Area of Science:

• Cognitive Science
• Neuroscience
• Motor Control

Background:

• Bayesian integration is a framework for perception, cognition, and motor control.
• Laboratory studies confirm Bayesian integration of sensory input and prior expectations for action guidance.
• Application to naturalistic tasks with complex movements and environmental statistics is less understood.

Purpose of the Study:

• To investigate if humans learn bimodal priors in a complex sensorimotor task: returning tennis serves.
• To determine if Bayesian principles extend to naturalistic, real-world scenarios.
• To explore implicit learning and exploitation of environmental statistics in motor behavior.

Main Methods:

• Participants performed a tennis serve return task in an extended reality setup.
• Serve locations followed a bimodal distribution, with visual uncertainty manipulated via ball speed.
• Behavioral data analyzed for biases related to prior distributions, uncertainty, and motor costs.

Main Results:

• Participant movements were systematically biased by the bimodal prior distribution after exposure.
• The magnitude of this bias correlated with visual uncertainty, consistent with Bayesian theory.
• Movements were also influenced by biomechanical constraints and motor costs.
• Participants implicitly used prior knowledge without explicit awareness of the serve pattern.

Conclusions:

• Humans can implicitly acquire and utilize bimodal priors in complex, naturalistic sensorimotor tasks.
• Bayesian learning extends beyond controlled lab settings to real-world activities.
• Motor behavior integrates prior expectations with sensory information, biomechanics, and motor costs, often unconsciously.