Influence of reward learning on visual attention and eye movements in a naturalistic environment: A virtual reality study

Abstract

Rewards constitute crucial signals that motivate approach behavior and facilitate the perceptual processing of objects associated with favorable outcomes in past encounters. Reward-related influences on perception and attention have been reliably observed in studies where a reward is paired with a unidimensional low-level visual feature, such as the color or orientation of a line in visual search tasks. However, our environment is drastically different and composed of multidimensional and changing visual features, encountered in complex and dynamic scenes. Here, we designed an immersive virtual reality (VR) experiment using a 4-frame CAVE system to investigate the impact of rewards on attentional orienting and gaze patterns in a naturalistic and ecological environment. Forty-one healthy participants explored a virtual forest and responded to targets appearing on either the left or right side of their path. To test for reward-induced biases in spatial orienting, targets on one side were associated with high reward, whereas those on the opposite side were paired with a low reward. Eye-movements recording showed that left-side high rewards led to subsequent increase of eye gaze fixations towards this side of the path, but no such asymmetry was found after exposure to right-sided high rewards. A milder spatial bias was also observed after left-side high rewards during subsequent exploration of a virtual castle yard, but not during route turn choices along the forest path. Our results indicate that reward-related influences on attention and behavior may be better learned in left than right space, in line with a right hemisphere dominance, and could generalize to another environment to some extent, but not to spatial choices in another decision task, suggesting some domain- or context-specificity. This proof-of-concept study also outlines the advantages and the possible drawbacks of the use of the 3D CAVE immersive platform for VR in neuroscience.

Fig 1. Illustration of the experimental setup.

Fig 2. Illustration of the timeline (top of the figure) and snapshot bullets depicting target events (bottom of the figure) during the VR experiment.

Participants had to explore a forest by walking along a path and find animals (rabbits or birds) that could appear on either the left or the right side with equal likelihood. Correct detection of animals on the left (Group Left_HR, left hemispace highly rewarded) or on the right side of space (Group Right_HR, right hemispace highly rewarded) was associated with a high reward (10 points), whereas correct detection of animals in the opposite hemispace was associated with a low reward (1 point) (snapshot bullets a) and b)). During this exploration phase, participants were also occasionally faced with path trifurcations (snapshot c)) and asked to choose to either go straight, turn 45 degrees to the left, or turn 45 degrees to the right. The forest was divided into three main sections (A, B, C), with a castle courtyard being displayed at the end of sections A and C. Participants were required to explore this courtyard and to find a hidden key (snapshot d)) (located in the opposite corner relative to the entrance in both castles). The second castle was presented with a 180 degrees rotation of its walls and internal layout.

Fig 3

Illustration of the courtyard layout for the two castles (Castle 1 at the end of section A, Castle 2 at the end of section C), for the Left_HR group (left panel) and for the Right_HR group (right panel). The key-target (surrounded by a yellow circle) was hidden at an unknown location, in the left opposite corner relative to the entrance for the Left_HR group and in the right opposite corner for the Right_HR group. The second castle was presented with a 180 degrees rotation. The key-target was presented in the same location in each castle.

Fig 4. Mean number of correct detections for left and right animal targets during sections A, B, and C for Left_HR participants (correct detection of animals on the left associated with a high reward) and Right_HR participants (correct detection of animals on the right associated with a high reward).

Fig 5. Mean number of gaze points directed towards the left / high-rewarded and the right / low-rewarded hemispace during sections A, B, and C for Left_HR participants (correct detection of animals on the left associated with a high reward) and Right_HR participants (correct detection of animals on the right associated with a high reward).

Fig 6

A. Illustrative heatmap depicting exploration trajectories of all Left_HR participants during their search in the courtyard, from their entrance until they found the hidden key. B. Group results for Left_HR, mean number of visited positions in the left and right parts of the courtyard in each castle. C. Group results for Right_HR, mean number of visited positions in the left and right parts of the courtyard in each castle.