The Relationship Between Environmental Statistics and Predictive Gaze Behaviour During a Manual Interception Task:...

Area of Science:

• Cognitive Science
• Neuroscience
• Computational Vision

Background:

• Human observers often exhibit Bayes-optimal decision-making.
• The visual system may utilize probabilistic mental models for environmental processing.
• Investigating eye movements in dynamic tasks can reveal underlying probabilistic computations.

Purpose of the Study:

• To determine if eye movements during a dynamic interception task align with Bayesian inference principles.
• To assess whether the gaze system adheres to environmental statistics and probabilistic models.
• To compare Bayesian learning models with associative learning models in explaining gaze dynamics.

Main Methods:

• Forty-one participants performed a virtual reality racquetball interception task.
• The probability of the ball's onset location was manipulated across five conditions.
• Gaze positions and trial-to-trial changes were analyzed to evaluate adherence to Bayesian principles.

Main Results:

• Pre-onset gaze positions tracked the true distribution of ball onset locations, indicating adherence to environmental statistics.
• Eye movements did not minimize distance to the fovea based on an optimal probabilistic model, instead reflecting a 'best guess'.
• Bayesian learning models (hierarchical Gaussian filter) better explained trial-to-trial gaze changes than associative learning models.
• Pupil dilations and gaze variability correlated with precision of beliefs and prediction errors from Bayesian models.

Conclusions:

• The gaze system appears to spontaneously adhere to environmental statistics during dynamic interception.
• While Bayesian learning influences gaze adjustments, eye movements may not fully optimize predictions in real-time.
• Probabilistic context significantly influences spontaneous gaze dynamics, as evidenced by pupil responses and gaze variability.