Christian B. Sinnott¹ (), Paul R. MacNeilage²; ¹Smith-Kettlewell Eye Research Institute, ²University of Nevada, Reno

Heading, the direction of linear self-motion in a head-centered reference frame, is estimated by the nervous system using vestibular and visual sensory cues. While accurate heading estimation is important for behaviors like locomotion (Cuturi and MacNeilage, 2013), human heading perception is biased, alternating between underestimation and overestimation of true stimulus values (Cuturi and MacNeilage, 2013; Crane, 2014). Perceptual biases can be modeled using Bayesian approaches that rely upon some representation of an organism’s previous experience, i.e., the prior distribution. The form of the prior can be constrained by measuring natural stimulus statistics, but to date there has been little work that has measured natural statistics of heading in humans. We therefore recorded head movements using a positional tracking camera (Hausamann et al., 2021) worn by ten participants over 50 hours of unconstrained, natural activity outside the lab. We use a kinematic calibration procedure detailed in previous work assessing human head orientation (Sinnott et al., 2023) to transform data out of sensor coordinates and into a head-centered reference frame. Positional tracking methods also allow for direct estimation of self-motion and heading direction. After pre-processing, we report data from approximately 42 hours of activity. Both heading azimuth (direction in the horizontal plane) and elevation (vertical plane) appear similar across participants, with means close to 0° (straight ahead); both azimuth and elevation exhibit high variability. To investigate this further we partitioned data into low (<0.75 m/s) and high (>0.75 m/s) speed epochs when participants predominantly performed stationary and locomotor tasks, respectively. While low-speed heading retains high variability, high-speed heading has decreased variability and high-speed heading azimuth appears Gaussian. We also present an early implementation of an efficient Bayesian model (Wei and Stocker, 2017) using our empirical measures of natural heading statistics to predict previously observed heading bias.

Acknowledgements: This research was supported by the NSF under Grant Number OIA-1920896 and NIH under Grant Number P20 GM103650