Quantification of Head Movement Predictability and Implications for Suppression of Vestibular Input during Locomotion.

Achieved motor movement can be estimated using both sensory and motor signals. The value of motor signals for estimating movement should depend critically on the stereotypy or predictability of the resulting actions. As predictability increases, motor signals become more reliable indicators of achieved movement, so weight attributed to sensory signals should decrease accordingly. Here we describe a method to quantify this predictability for head movement during human locomotion by measuring head motion with an inertial measurement unit (IMU), and calculating the variance explained by the mean movement over one stride, i.e., a metric similar to the coefficient of determination. Predictability exhibits differences across activities, being most predictable during running, and changes over the course of a stride, being least predictable around the time of heel-strike and toe-off. In addition to quantifying predictability, we relate this metric to sensory-motor weighting via a statistically optimal model based on two key assumptions: (1) average head movement provides a conservative estimate of the efference copy prediction, and (2) noise on sensory signals scales with signal magnitude. The model suggests that differences in predictability should lead to changes in the weight attributed to vestibular sensory signals for estimating head movement. In agreement with the model, prior research reports that vestibular perturbations have greatest impact at the time points and during activities where high vestibular weight is predicted. Thus, we propose a unified explanation for time-and activity-dependent modulation of vestibular effects that was lacking previously. Furthermore, the proposed predictability metric constitutes a convenient general method for quantifying any kind of kinematic variability. The probabilistic model is also general; it applies to any situation in which achieved movement is estimated from both motor signals and zero-mean sensory signals with signal-dependent noise.