Depth perception changes following adaptation to cue-dependent invariants.

How does our perceptual system adapt to new invariants? Can the visual system adapt to non-veridical 3D object properties that remain stable under different transformations? To investigate this, we employed a two-cue depth stimulation paradigm where disparity and texture were in conflict. We found that active visuomotor interaction with a metameric 3D planar surface (i.e., self manipulation of a metameric 3D planar surface) drives adaptation, altering the perceived match of different combinations of stereo and texture information. Notably, this adaptation occurred through action-driven exposure to visual invariants alone, without the need for explicit sensorimotor error feedback. Adopting a 3D vector-sum model to jointly account for both slant and tilt weighting contributions, we analyzed the effects of dynamic and active training on cue integration. By comparing pre- and post-training perceptual judgments, we found that such training induces a post-training cue reweighting, favoring one cue over the other. Control experiments confirmed that this effect requires the dynamic, coherent manipulation of both cues since no significant adaptation occurred when motor action was yoked to a single cue in isolation. Overall, these results demonstrate that new, coherent precepts emerge when sensory cues are dynamically manipulated in a coherent manner, thereby adapting cue-integration mechanisms to new high-level structural invariants. Our work underscores that active exploration is a crucial mechanism for perceptual learning, facilitating the grouping of features defined by systematic (i.e., invariant) relationships among structural properties of the visual signal (in our case the amount of the cue conflict) thus enabling adaptation to novel environmental statistics.