Mice navigate complex landscapes by learning a sparse vocabulary of paths

Ma, T.; Comrie, A. E.; Pjanovic, V.; Karashchuk, L.; Dennis, E. J.; Hermundstad, A. M.; Voigts, J.

doi:10.64898/2026.10.10.757726

Home/Archive/bioRxiv

bioRxiv · 10.64898/2026.10.10.757726

Mice navigate complex landscapes by learning a sparse vocabulary of paths

Ma, T.·Comrie, A. E.·Pjanovic, V.·Karashchuk, L.·Dennis, E. J.·Hermundstad, A. M.·Voigts, J.

View original source ↗Export metadata ↓

Abstract

Effective navigation is critical for survival, but how animals contend with the scale, complexity, and self-similarity of natural landscapes is unknown. Here, we investigate spontaneous exploratory mouse behavior, both outdoors and in dark self-similar landscapes made from tens to hundreds of tiles of varying heights. Within 2 hours of exploration, mice detected small covert changes to the landscape and jumped across ledges to invisible landing spots, revealing detailed spatial knowledge. In this process, mice converged on a sparse set of memorized paths. These routes depended on prior experience and increased the compressibility of sensory information over time. Our results demonstrate that mice rapidly learn sparse internal representations of their surroundings and suggest that animals use structured behavioral sequences to reduce the complexity of natural navigation.

Explore related subjects

animal behavior and cognition Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Ma, T., Comrie, A. E., Pjanovic, V., Karashchuk, L., Dennis, E. J., Hermundstad, A. M., Voigts, J.. 2026-10-11. Mice navigate complex landscapes by learning a sparse vocabulary of paths. https://doi.org/10.64898/2026.10.10.757726

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Viewing geometry and pattern type shape the detectability of snakes

Protective coloration can reduce predation risk by altering how prey are detected, recognised, or attacked. High-contrast markings may function as warning signals, camouflage, or both. In snakes, dorsal colour patterns are highly variable but the function remains debated, particularly because their appearance to an observer may change with viewing distance and angle. This study tested whether the function and efficacy of common snake pattern types (blotches, spots, diamonds, zig-zags and bands) is observer-view dependent, mediated by viewing distance and angle. Artificial snake models marked with these different pattern types that are broadly common across snake taxa, were placed in woodland habitat with a mixed shrub and grass understorey in Cornwall, UK, with photographs taken for evaluating levels of predicted camouflage and contrast against the background. Next, human volunteers searched for models, and we recorded detection distances and angles. Field results showed no significant difference in overall detection rates between pattern types; however, zigzag-patterned models were more likely to be detected at significantly shorter distances than blotch-patterned models, with a marginal trend towards shorter distances than diamond-patterned models, indicating effective longer-range camouflage. Detection risk did not alter gradually as volunteers approached but increased abruptly, correlating with image analyses showing that levels of predicted disruptive camouflage decreased with reduced distance. When detection was analysed using viewing angle, an advantage of zigzag patterns was robust to geometric scale, and observer height may further shape detection through viewing geometry. These findings suggest that snake dorsal patterns could potentially serve dual roles in reducing detection from afar, while being more conspicuous at close range. In addition, the results highlight the need to consider three-dimensional viewing geometry, including distance and visual angle, when evaluating the ecological function of animal coloration.

animal behavior and cognition↗

bioRxiv2026

Camouflage functions of juvenile giant panda colouration in forest environments

The contrasting black-and-white coloration of the giant panda (Ailuropoda melanoleuca) has long attracted attention. Recent works suggest this contrasting pattern may generate camouflage-relevant visual effects. However, most previous studies have focused on adult pandas, whereas juveniles may face greater predation risk and may therefore represent a more ecologically relevant life stage for testing concealment hypotheses. We investigated whether juvenile panda coloration provides the potential for camouflage through background matching or disruptive coloration, and how this potential is shaped by vegetation cover, illumination and background position. We used life-sized juvenile panda models with natural and experimentally modified coloration by adjusting the relative proportions and spatial arrangement of white, black, and grey patches. Models were photographed in temperate woodland environments at ground level and against canopy backgrounds, under sunny and cloudy conditions, with and without vegetation occlusion. We quantified colour and luminance contrast, as well as edge-disruption metrics, using dichromatic canid and felid visual models. Natural juvenile panda coloration influences target-background contrast and edge-disruption, although these effects varies with vegetation, lighting, background height and visual model. Vegetation cover significantly reduced contrast and increased edge disruption metrics, while bright light condition only increased the metrics of edge disruption. Background position affected camouflage-relevant metrics in a context- and observer-dependent manner: under vegetation occlusion, canopy placement reduced chromatic contrast under felid-like vision, whereas ground-level placement reduced luminance contrast under both visual models. Alterations to patch extent and boundary arrangement modified camouflage-relevant metrics under specific environmental conditions, suggesting that the natural arrangement of pelage boundaries may enhance concealment under specific conditions. Overall, our findings support the hypothesis that juvenile giant panda coloration can produce visual effects consistent with potential camouflage, while highlighting the importance of habitat structure and viewing context in shaping camouflage.

animal behavior and cognition↗

bioRxiv2026

From Wandering Thoughts to Mental Stillness: A Minimal Computational Statistical-Mechanical Model of Driven Relaxation

Reduced spontaneous thought during meditation is often described phenomenologically, but a mechanistic account should specify how a dynamical system can enter and persist in such a regime. Here we formulate a minimal statistical-mechanical model in which a dimensionless coordinate, q, represents a coarse-grained propensity for spontaneous-thought activity, and a slowly relaxing variable, r of t, represents the history-dependent effect of an attentional manipulation. The model combines a tilted double-well landscape with Langevin noise, allowing a high-q and a low-q dynamical regime to coexist. We numerically integrate the model under a controlled focusing protocol followed by removal of the drive and quantify low-state occupancy, first-passage times, dwell times, occupancy entropy, and transition entropy. For the chosen dimensionless parameterization, increasing the drive amplitude produces a sharp crossover in low-state occupancy near the deterministic spinodal threshold, A-sub-sp, approximately equal to 0.0148. At a noise intensity of D equal to 0.01, the probability of occupying the low-q regime at drive removal increases from 0.094 at A equal to zero to 0.994 at A equal to 0.02, while the mean first-passage time decreases from 48.0 to 13.5 model time units. Noise produces a non-monotonic effect: for A equal to 0.01, low-state occupancy at drive removal is only 0.033 at D equal to 0.0025, rises to 0.842 at D equal to 0.02, and falls to 0.718 at D equal to 0.04. Strong driving also reduces the effective number of occupied states while changing transition entropy much less, illustrating why a single entropy measure cannot identify thoughtlessness. These results do not establish that human neural dynamics obey this model. Instead, they show that a driven stochastic bistable system can generate the qualitative sequence of focusing, switching, relaxation, and metastable low-activity residence, and they provide quantitative predictions that can be tested against time-resolved neural recordings and independent reports of subjective experience.

animal behavior and cognition↗