From simple lab tasks to the virtual court: Bayesian integration in tennis

Area of Science:

• Cognitive Neuroscience
• Motor Control
• Human-Computer Interaction

Background:

• Human behavior often relies on integrating sensory information with prior expectations, a process modeled by Bayesian inference.
• Existing evidence for Bayesian integration is primarily from simplified laboratory tasks, limiting understanding of its application in complex, real-world scenarios.

Purpose of the Study:

• To investigate whether human predictive gaze behavior in a complex sensorimotor task, returning tennis serves, aligns with Bayesian integration principles.
• To examine how prior expectations and sensory evidence are dynamically weighted based on reliability over different timescales.

Main Methods:

• Two experiments were conducted using an extended reality (XR) setup simulating real tennis serve returns with unconstrained movements.
• Participants (n=64) faced opponents with distinct serve location distributions, and their predictive gaze behavior and explicit judgments of ball-bounce location were recorded.
• Visual uncertainty was manipulated by varying ball speeds in the second experiment.

Main Results:

• Predictive gaze behavior demonstrated a bias towards an opponent's preferred serve locations, consistent with Bayesian predictions.
• This prior-driven bias was more pronounced under increased visual uncertainty (higher ball speeds).
• Reliability-weighted integration was observed on two timescales: prior influence increased with match experience, while single-serve estimates were updated by incoming sensory data.

Conclusions:

• Bayesian theory offers a robust framework for understanding sensorimotor adaptation in complex, high-performance tasks.
• Extended reality technology enables testing of real-world motor control principles under controlled experimental conditions.
• The findings highlight dynamic, reliability-based integration of prior knowledge and sensory input in guiding human behavior.