The Relationship Between Environmental Statistics and Predictive Gaze Behaviour During a Manual Interception Task: Eye Movements as Active Inference

Abstract

Human observers are known to frequently act like Bayes-optimal decision-makers. Growing evidence indicates that the deployment of the visual system may similarly be driven by probabilistic mental models of the environment. We tested whether eye movements during a dynamic interception task were indeed optimised according to Bayesian inference principles. Forty-one participants intercepted oncoming balls in a virtual reality racquetball task across five counterbalanced conditions in which the relative probability of the ball’s onset location was manipulated. Analysis of pre-onset gaze positions indicated that eye position tracked the true distribution of onset location, suggesting that the gaze system spontaneously adhered to environmental statistics. Eye movements did not, however, seek to minimise the distance between the target and foveal vision according to an optimal probabilistic model of the world and instead often reflected a ‘best guess’ about onset location. Trial-to-trial changes in gaze position were, however, found to be better explained by Bayesian learning models (hierarchical Gaussian filter) than associative learning models. Additionally, parameters relating to the precision of beliefs and prediction errors extracted from the participant-wise models were related to both task-evoked pupil dilations and variability in gaze positions, providing further evidence that probabilistic context was reflected in spontaneous gaze dynamics.

Counteracting Uncertainty: Exploring the Impact of Anxiety on Predictive Sensorimotor Control Mechanisms in a Manual Interception Task

“…Eye tracking data was lost for one participant in one condition due to a recording error. In a previous study, we observed effects in the range of R 2 = .03-.06 for changes in predictive gaze position in this task (44). Power curves using the simr package for R (45) showed that for linear mixed effects models (with participant as a random factor), 36 trials for 30 participants was sufficient to detect effects at the lower end of this range with more than 85% power (see power curves in the supplementary files: https://osf.io/2amdv/).…”

“…The virtual environment, which had the visual appearance of an indoor racquetball court, was developed using the gaming engine Unity (v2019.3.1f; Unity Technologies, San Francisco, CA) (see Fig 1). The same task has been used to study predictive eye movements in previous papers (44,46). The task was displayed using an HTC Vive head-mounted display (HTC Inc., Taoyuan City, Taiwan).…”

“…Predictive gaze location, was defined as the horizontal (xcoordinate) gaze position (i.e., gaze vector intersection point with the world) at the termination of the auditory tones, averaged over a 50ms window. Following Harris et al (44) this was taken as the most critical moment for anticipating the projection location and therefore the critical eye movement feature that would reveal participants' prediction about the upcoming trial. The predictive gaze location variable was then used as an input to model the updating of sensorimotor predictions over trials (described in the Active Inference Modelling section below).…”

“…The purpose of this model was to use the observed behaviour (eye position) to infer an unobserved variable (belief about the next release location). We therefore chose a gaze variable (position before release) that had a clear mapping to a task-relevant belief (release location) (as in (44)). A Bayesian generative model of perception derived from the Markov decision process (MDP) formulation of active inference (61) was chosen to align with the theoretical approach of the paper.…”