Statistical determinants of visuomotor adaptation in a virtual reality three-dimensional environment

Abstract

AbstractRecent advances in virtual reality (VR) technology allow motor control studies to investigate movement variety in a controlled 3D environment. However, visuomotor adaptation paradigms for arm reaches in a 3D space are still scarcely investigated. Elucidating these adaptations is important to further our understanding, predicting and hence, rehabilitating everyday movement behavior in humans suffering from motor deficits. We test how well optimal integration theory that was extrapolated from 1D and 2D settings extends to movements in 3D space; specifically, if and how a subject’s variability in planning and sensory perception will influence motor adaptation. We test this hypothesis without introducing artificial (experimentally induced) variability. Instead, we exploit the natural anisotropy in sensory performance along different dimensions relative to the body. We use a virtual reality (VR) setup to compare the subject’s adaptation of their 3D movements to perturbations applied separately in sagittal, coronal and horizontal planes. We found that subjects adapt best to perturbations in the coronal plane, i.e. when the depth component of the movement is unaffected, followed by sagittal then horizontal perturbations. To characterize how the adaptation rate selectively depends on planning and sensory uncertainty along different body axes, we developed a novel hierarchical two-state space model using Bayesian modeling via Hamiltonian Monte Carlo procedures to fit visuomotor adaptation profiles. Our hierarchical model reconstructs learning parameters from surrogate data better than a state-of-the-art expectation-maximization (EM) algorithm. Modeling revealed that differences in adaptation rate of movement among the three planes are explained from the Kalman gain, i.e. the relative variability associated with planning and sensory perception. The developed method allowed us to show that the motor behavior is in line with optimal integration theory when considering adaptation rate variation between the different planes.Author SummaryThe process of neurorehabilitation in human patients suffering from motor deficits could be considered as a process of relearning or re-adapting motor skills. Virtual reality is already used to drive motor rehabilitation. However, the majority of motor learning theories were developed under one or two-dimensional experimental paradigms and it is assumed that the general findings apply to motor control in three-dimensional movements. To test whether these theories are also compatible with more naturalistic movements performed in a three-dimensional space, we used a new statistical method to identify the dynamics of motor adaptation in virtual reality. We found that in a three-dimensional environment, there are differences in how the subject will correct and adapt their movement based on which orientation relative to the body the movement needs to be corrected. These differences come from the statistical variability associated with unperturbed movements in the different directions. These findings provide clues on how therapeutic regimens in virtual reality should be adapted, i.e. physical reorientation of the body, to improve rehabilitation of certain movements.