The Computational Mechanics of the Cognitive Map

67 concepts47 papers1 proposal

To navigate an unpredictable world, an animal must maintain an internal representation of its position and orientation—a biological "GPS" known as the cognitive map. For decades, neuroscience has identified the cellular building blocks of this map: place cells in the hippocampus that fire at specific, discrete locations, and grid cells in the medial entorhinal cortex (MEC) that fire in a regular, hexagonal lattice across an environment. Clinical evidence underscores the necessity of this system; damage to the hippocampus profoundly impairs allocentric spatial navigation and episodic memory while leaving basic visual pattern matching intact, as demonstrated by Hugo J. Spiers et al. in Bilateral hippocampal pathology impairs topographical and episodic memory but not visual pattern matching.

Yet, knowing what these cells are does not explain why they take these specific geometric forms, nor how they coordinate to maintain a stable map. Explaining these properties remains a central challenge in spatial cognition (Caswell Barry in Neural Mechanisms of Self-Location). Recent computational and theoretical work reveals that the cognitive map is not a rigid, hardwired hierarchy, but an emergent, mathematically optimal solution to spatial inference driven by joint network dynamics.

Spatial Representations as Mathematical Optimality

Why hexagons? Why discrete place fields? Instead of being biological accidents, these patterns emerge naturally when neural networks are optimized to solve basic navigation constraints.

Consider a recurrent neural network (RNN) trained to perform path integration—updating its estimated position using only velocity inputs. As shown by Christopher J. Cueva et al. in Emergence of grid-like representations by training recurrent neural networks to perform spatial localization and the authors of Training recurrent networks to generate hypotheses about how the brain solves hard navigation problems, these networks spontaneously develop internal representations that mirror biological grid cells and border cells. The hexagonal grid pattern is simply the mathematically optimal way to compress and represent two-dimensional self-motion information within a resource-constrained recurrent network.

Similarly, if we apply unsupervised learning principles to raw visual streams, the statistics of an animal's movement dictate what representations emerge. Mathias Franzius et al. demonstrated in Slowness and Sparseness Lead to Place, Head-Direction, and Spatial-View Cells that a single framework optimizing for "slowness" (the principle that spatial location changes continuously and slowly over time) and "sparseness" naturally generates place cells, head-direction cells, and spatial-view cells from visual inputs.

Beyond Hierarchy: Joint Attractor Dynamics

The classic textbook view suggested a simple feedforward hierarchy: grid cells compute distances from self-motion and project to the hippocampus to form place cells. However, this view fails to capture the reciprocal nature of the system.

In reality, the hippocampus and entorhinal cortex form a tightly coupled loop governed by joint attractor dynamics, as proposed by Yoram Burak et al. in A theory of joint attractor dynamics in the hippocampus and the entorhinal cortex accounts for artificial remapping and grid cell field-to-field variability. Rather than grid cells strictly dictating place cell firing, place cells can emerge and remap using environmental boundary information alone, bypassing the need for grid cell inputs (Daniel Bush et al. in What do grid cells contribute to place cell firing?; Markus Pettersen et al. in Learning place cells and remapping by decoding the cognitive map).

Conversely, when the hippocampus is inactivated and the spatial grid pattern vanishes, the underlying network of grid cells still maintains its internal firing synchrony, as shown by Noam Almog et al. in During hippocampal inactivation, grid cells maintain synchrony, even when the grid pattern is lost. This proves that the entorhinal cortex possesses its own intrinsic recurrent network structure.

   +-----------------------+                  +-----------------------+
   |      Hippocampus      |  Joint Attractor |   Entorhinal Cortex   |
   |     (Place Cells)     | <============--> |     (Grid Cells)      |
   +-----------------------+     Dynamics     +-----------------------+
               ^                                          ^
               |                                          |
       Boundary Inputs                             Self-Motion / Cues

This dual-system architecture allows the brain to solve a fundamental challenge: integrating internal self-motion (path integration) with external sensory landmarks. As modeled by the authors of Neural Dynamics Indicate Parallel Integration of Environmental and Self-Motion Information by Place and Grid Cells, place and grid networks act as parallel attractors. When an animal encounters a conflict—such as a landmark moving unexpectedly—the joint system smoothly resolves the discrepancy, utilizing specialized "cue cells" in the MEC to correct path integration drift (Amina A. Kinkhabwala et al. in Visual cue-related activity of cells in the medial entorhinal cortex during navigation in virtual reality).

Balancing Accuracy, Robustness, and Scale

If grid codes are highly efficient for calculating distance, why does the brain also maintain complex, multi-scale place codes where single cells fire in multiple locations? Robin Dietrich explored this trade-off in Grid codes vs. multi-scale, multi-field place codes for space, revealing that while uniform grid codes provide the highest absolute positional accuracy, multi-field place codes offer vastly superior robustness against noise and physical brain lesions.

Furthermore, the brain does not rely on a single monolithic map, but a manifold of interconnected spatial maps (Edvard I Moser et al. in A manifold of spatial maps in the brain). To prevent these maps from becoming unmanageably large in complex environments, the hippocampal-entorhinal circuit leverages compositionality. Christopher Kymn et al. proposed in Binding in hippocampal-entorhinal circuits enables compositionality in cognitive maps that the brain uses an algebraic framework akin to a residue number system combined with vector binding operations. This allows the network to compose high-resolution, noise-robust representations of large spaces out of simpler, low-dimensional neural building blocks.

This rich local encoding is reflected in the fact that grid cell fields are not perfectly uniform; they exhibit stable, field-to-field rate variability that embeds local positional context directly within the global metric grid (Revekka Ismakov et al. in Grid Cells Encode Local Positional Information).

Dynamic Inference and Temporal Coordination

A cognitive map is not just a static blueprint; it is a dynamic engine for inference and planning. The brain achieves this by organizing spatial representations in time through neural oscillations.

As an animal moves through a place field, the corresponding place cells fire at progressively earlier phases of the local theta rhythm—a phenomenon known as phase precession (Neil Burgess in Oscillations and Boundaries in My Route Through the Hippocampal Cognitive Map). Tianhao Chu et al. demonstrated in Firing rate adaptation affords place cell theta sweeps, phase precession, and procession that this temporal sequencing, along with forward and backward "theta sweeps," can emerge naturally from simple firing rate adaptation within the neural network.

These temporal sweeps allow the brain to look ahead and behind, transforming a spatial map into a predictive simulator. By framing spatial cognition as a probabilistic system executing forward and backward inference, as detailed by W.D. Penny et al. in Forward and Backward Inference in Spatial Cognition, the brain can plan paths, evaluate choices, and consolidate memories. When this delicate coordination breaks down—such as the reduced grid-like theta modulation observed by Laura Convertino et al. in Reduced grid-like theta modulation in schizophrenia—the ability to form stable relational and spatial representations is profoundly impaired.

Key Foundational Work

Using hippocampal 'place cells' for navigation, exploiting phase coding (1992) — contributed 1 concept elements to this direction.
Simulation of a Thalamocortical Circuit for Computing Directional Heading in the Rat (1995) — contributed 1 concept elements to this direction.
Using Hippocampal 'Plane Cells' for Navigation, Exploiting Phase Coding (1992) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_neural_basis_of_spatial_navigation_d287b6bc,
  title = {neural basis of spatial navigation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/neural-basis-of-spatial-navigation/rdr_d287b6bc7e724b2ea6571bb8e75163db},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_neural_basis_of_spatial_navigation_d287b6bc_v1,
  title = {neural basis of spatial navigation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/neural-basis-of-spatial-navigation/rdr_d287b6bc7e724b2ea6571bb8e75163db/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake