Pairs that coordinated better than expected from their individual skill made fewer corrective eye movements during a joint-action task, researchers report. The difference appeared in both roles, the hitter and the receiver, and in both phases of the exchange. The result points to a relationship between successful joint action and the way partners anticipate and adjust their gaze, not to proof that one caused the other.
The study was built around a question: does coordination depend mainly on online sensory extrapolation, internal predictions built from prior action, or a combination of both? To make the comparison, researchers estimated expected joint performance from each pair's mean individual-task score using a Bayesian model. Pairs whose joint performance landed above that estimate were labeled above-expected, while the rest were labeled below-expected. This definition accounted for baseline skill before the individual predictors were evaluated.
A comparison built around expected performance
The study recruited 91 participants aged 19 to 45. The group included 36 female participants, 54 male participants and one non-binary participant. The mean age was 25.29 years, with a standard deviation of 7.09 years. After exclusions based on eye-tracking quality and total task score, nine of 91 individual-action tasks and 15 of 79 joint-action tasks were removed, leaving 70 participants and 64 valid dyads.
Participants completed an individual-action task lasting approximately three minutes and a joint-action task lasting five to six minutes, with the order counterbalanced. Mean event success was 90.82% in the individual task and 85.21% in the joint task.
Two eye signals did most of the sorting
To find signals that separated the two groups, the researchers used XGBoost, a machine-learning classifier. It considered individual-task eye-movement, motor, heart-rate, movement-energy and static predictors. The two strongest predictors were the dissimilarity between partners in visual-angle error and the mean number of corrective saccades in Phase 1, meaning quick eye movements used to adjust gaze.
Using only those two signals, the model classified pairs with 75% accuracy. The 95% confidence interval ran from 73.66% to 76.34%, giving a statistical range around that estimate. The analysis reported that this two-signal model outperformed the full model by 4.69%. The figure describes classification performance, not how much better one pair played than another.
Better pairs corrected less
Above-expected pairs made fewer corrective saccades than below-expected pairs in both the hitter and receiver roles and in both phases. The median effect sizes were -0.58 for hitters in Phase 1, -0.47 for hitters in Phase 2, -0.45 for receivers in Phase 1 and -0.48 for receivers in Phase 2. The reported 95% highest-density intervals, the study's uncertainty ranges, all remained below zero.
The differences were not limited to the number of corrections. Above-expected hitters made smaller and more precise anticipatory saccades in Phase 1. Receivers made larger Phase-1 saccades. In Phase 2, both roles initiated their saccades later, while receivers' movements were less precise. The better-coordinated pairs therefore showed a different pattern of gaze size, accuracy and timing.
A finding with clear boundaries
Taken together, the findings are consistent with a simple interpretation: coordination was better when partners' visual tracking errors and gaze corrections were more compatible. That is an interpretation of an association, not proof that one eye movement caused success. The evidence came from a controlled task lasting about three minutes for the individual condition and five to six minutes for the joint condition, with 70 participants and 64 valid pairs in the final dataset. The result should therefore be read as evidence about this setup, rather than a general rule for every form of cooperation.
Paper data and sources
Original title: Successful coordination emerges from aligned self-predictions in a dynamic motor interaction.
Authors: Prasetia Utama Putra, Fumihiro Kano
Journal/Repository: Communications psychology
Status: Corrected publication
First online: 2026-08-20
DOI: 10.1038/s44271-026-00523-7
Original paper · Full text