Grounded expectations: stability of sensorimotor priors during vertical pointing in a virtual environment

Humans rely on well-calibrated internal models of physical laws, such as gravity, to guide efficient manual actions. In this study, we investigated whether such gravitational expectations are altered in virtual reality (VR) and how this might influence the execution and adaptation of goal-directed pointing movements. We compared pointing movements in physical and virtual environments, focusing on initial acceleration as an index of feedforward control. To capture trial-by-trial adaptation and the influence of prior beliefs on pointing movements, we modeled this data using the generalized hierarchical Gaussian filter, a Bayesian computational model of learning under uncertainty. Initial hand acceleration was found to be slightly lower in the virtual environment than in the physical environment, but no condition-related difference was found in variability of acceleration. Model-estimated gravity beliefs were found to be similar between virtual and physical environments, but belief certainty was observed to decline across trials in the virtual condition, suggesting an accumulation of uncertainty over time. In summary, gravity priors remained stable in VR, guiding action similarly to physical environments, but the sensory uncertainty of VR eroded the precision of these priors over time.<b>NEW & NOTEWORTHY</b> This study demonstrates that humans' sensorimotor priors about gravity remain stable when performing vertical pointing movements in virtual environments, despite accumulating sensory uncertainty over time. Using kinematic measures and a Bayesian computational model, we show that core predictive control transfers from real-world to immersive contexts but confidence in predictions declines with prolonged virtual reality (VR) exposure. These findings advance understanding of how predictive motor control adapts to VR, with implications for training, rehabilitation, and human-computer interaction.