Scene variability affects action decisions, confidence and behaviour dynamics.

Humans often unfold their actions within complex, variable environments. While such variability is sometimes irrelevant to task success, it could still influence both action decisions and confidence in their outcome. We used immersive virtual reality to examine how scene variability affects each of these processes, while also testing whether its influence extends to the dynamics of information sampling and motor execution. Our task simulated a street-crossing scenario where participants saw six cars approaching a crosswalk. In separate conditions, participants judged their confidence in crossing successfully or attempted to cross if deciding so. Across trials, scene variability was manipulated by changing the cars' relative velocities. Higher scene variability led participants to report lower confidence and to attempt crossing less often. Moreover, it delayed crossing initiation, which participants compensated for by moving slightly faster once committed to the action. These kinematic adjustments indicate flexible online control in response to environmental variability. Consistent with this interpretation, overall crossing performance remained high across variability conditions, suggesting that such adjustments enabled participants to maintain successful outcomes despite increased scene complexity. Additionally, eye movement patterns revealed that trials with high confidence and attempted crossings were associated with faster information acquisition engagement and subsequent disengagement. Our findings suggest that, even when irrelevant for action success, variability can shape not only confidence and action decisions but also the dynamics of motor execution. This highlights the importance of accounting for variability to fully understand sensorimotor tasks in dynamic complex environments.