Fig. 1: Macaques navigate in virtual reality to latent targets using path integration and an acquired internal model mapping joystick position to self-motion velocity.

A Setup. Monkeys use a joystick controlling their linear (max = 200 cm/s) and angular (max = 90 deg/s) velocity to navigate to the location of a briefly (300 ms) presented target. Monkeys were trained with a high-density of ground-plane elements, thus having access to a self-velocity signal (motion across their retina) they integrated into an evolving estimate of position. During recordings, we additionally used a low-density condition, wherein monkeys had to primarily rely on an internal model mapping joystick position to virtual velocity (akin to “stride size”). Targets were presented within a radial (1–4 m) and angular (−45 to 45 deg relative to straight ahead) range, and we quantified performance by regressing radial (r) and angular eccentricity (\({{{{{\rm{\theta }}}}}}\)) of responses and targets. Figure adapted from REF²⁶ B Example session. Monkeys tended to undershoot targets in radial distance (top) and eccentricity (bottom). For each session, we fit a linear regression (response = gain x target location + intercept) and thus a gain of ‘1’ indicates no bias. Gains <1 indicate undershooting. C Radial and angular biases during high- and low-density conditions. Q, S, and M refer to three individual monkeys. Transparent small circles are individual sessions, and the larger opaque blue circle is the mean for each animal. The rightmost circle within each panel is the mean across animals, with error bars indicating ± 1 S.E.M. D Endpoint error and variance. Scatter plot of endpoint errors along the depth (y-axis) and lateral (x-axis) dimension as a function of density. For visualization, data is pooled across all sessions and monkeys (high-density n = 44,476 trials; low-density n = 43,286 trials). Isolines showing two dimensional histograms demonstrate the increased spread (variance) of endpoints relative to target in the low-density condition. Source data are provided as a Source data file.