Abstract
Humans can determine their heading accurately from optic flow. However, previous studies have shown that heading directions are sometimes underestimated (e.g., Sun et al. 2020) yet other times overestimated (e.g., Cuturi/MacNeilage, 2013) despite using similar optic flow stimuli. Here we show that these contrasting findings do not reflect actual differences in heading perception but are caused by experimental differences in how participants reported their estimates. We ran a psychophysical experiment where participants were virtually translating in a 3D dot-cloud space with different heading directions. On each trial, an optic flow display was presented indicating a heading direction uniformly sampled from a range of +/-33 deg. After stimulus presentation, participants reported their perceived headings by adjusting a probe on a circle. Three conditions were tested that only differed in the size of the response range within which participants were able to report their estimates (80, 160, 240 deg on the circle). We found that heading estimates were proportionally scaled with the size of the response range such that they were overestimated in large range conditions but underestimated in the smallest range condition. We also derived a Bayesian observer model to quantitatively characterize participants’ estimation behavior. The model assumes efficient sensory encoding that reflects the neural coding accuracy in area MSTd (Gu et al. 2010). Furthermore, it assumes estimates that are linearly scaled for each range condition. We found that this model quantitatively well predicted participants’ estimates both in terms of mean and variance. Our results imply that participants heading percepts are identical under all three response conditions and are well explained with a Bayesian observer model. The differences in reported estimates can be solely attributed to different sizes of the response range and are fully explained by a linear mapping from the percept to the probe response.