Spatial Navigation

Abstract

Spatial navigation is a type of mental process: the operation by which an animal tracks its position and heading and computes a route to a goal. This article sets out its two reference frames — egocentric, anchored to the body, and allocentric, anchored to the world — and the cognitive map that Tolman inferred from how rats learn a maze. It then traces the neural machinery the map proved to have: place cells in the hippocampus, grid cells that tile space in a hexagonal lattice, and head-direction cells acting as a neural compass. It follows this code into the human brain, underlying navigational expertise, a grid-like fMRI signal, and an early marker of Alzheimer's disease. Three demonstrations let the reader switch reference frames, build a grid pattern, and integrate a path into a homing vector.

Keywords: cognitive map, place cells, grid cells, path integration, allocentric

An animal that leaves its nest to forage and returns by a straight line has solved a problem no single sense hands it ready-made. Neither the eye nor the ear reports where home is; that must be computed, held, and continuously updated as the animal moves. Spatial navigation is the family of mental processes that does this computing — the estimation of one's own position and heading, the representation of where places and goals lie, and the planning of a route between them (Epstein, Patai, Julian, & Spiers, 2017). It is at once one of the oldest problems a nervous system faces and, since the discovery that the mammalian brain contains cells tuned to place, direction, and distance, one of the best understood links between a cognitive function and its neural implementation. This article follows that link from the behavioural evidence for an internal map, through the cell types that compose it, to its expression and its failure in the human brain.

What Spatial Navigation Is

Spatial navigation is the process of determining and maintaining a course from one place to another. It decomposes into a few interlocking sub-problems: localisation, knowing where one currently is; orientation, knowing which way one is facing; and wayfinding, computing and following a route to a goal that may be out of sight. What makes it a cognitive problem rather than a reflexive one is that the information needed is rarely present in the immediate stimulus. A goal behind the animal casts no image on its retina; the distance already travelled leaves no mark on the world. The position and heading that navigation trades in must be represented internally and kept current as the body moves, which is why the study of navigation has always been, at bottom, the study of an internal model of space.

The founding demonstration that animals build such a model came from Edward Tolman. Behaviourism held that a rat learns a maze as a chain of stimulus-response associations — at this junction turn left, at that one turn right — each stamped in by reward. Tolman's latent-learning experiments broke that account. Rats allowed to explore a maze with no reward, and rewarded only later, learned the goal's location almost immediately once reward was introduced, as though the unrewarded wandering had built knowledge that lay latent until it became useful. Rats also took novel shortcuts and detours that no reinforced turn could explain. Tolman concluded that the animal had acquired not a set of responses but a cognitive map — an internal representation of the spatial layout that it could read to compute routes it had never practised (Tolman, 1948). The claim was radical in 1948 precisely because it posited an unobservable internal representation, the very thing behaviourism had ruled out, and it became a founding case for the cognitive revolution.

Egocentric and Allocentric Frames

To represent where something is, a system must choose an origin and axes — a reference frame. Navigation uses two, and the difference between them organises much of the field. An egocentric frame places the origin on the navigator's own body and describes locations by their bearing and distance relative to it: the café is thirty metres ahead and to my left. Egocentric representations are immediate and are what action needs — to reach for or walk toward a thing, the brain must know where it is relative to the body — but they are fragile, because every one of them is invalidated the moment the body turns or moves. An allocentric frame places the origin in the world and describes locations by their relations to each other and to fixed landmarks, independent of where the navigator happens to be standing: the café is on the north side of the square, next to the fountain. Allocentric representations are stable across the navigator's movements and support flexible route-planning, shortcutting, and perspective-taking, and it is an allocentric representation that the term cognitive map denotes (Burgess, Maguire, & O'Keefe, 2002).

How a navigator comes to possess an allocentric map is itself a developmental progression. In an influential synthesis, Alexander Siegel and Sheldon White argued that knowledge of a large-scale environment is acquired in a fixed order: first landmark knowledge, the recognition of distinctive individual places; then route knowledge, the sequences of landmarks and turns that link them, still tied to the traveller's own path; and finally survey knowledge, an integrated, map-like representation of how all the places lie relative to one another, from which shortcuts and novel routes can be read off (Siegel & White, 1975). The endpoint of that sequence is precisely an allocentric cognitive map, while the earlier stages are more egocentric and route-bound — which is why survey knowledge is the hardest-won, and, as the clinical section returns to, among the first casualties when the mapping system begins to fail.

The two frames are not rivals but a division of labour, and navigation continually translates between them. A stable allocentric map is useless for action until it is converted into an egocentric command — turn left now — and the egocentric stream of self-motion must be folded back into the allocentric map to keep it registered to the world. Figure 1 shows the three canonical cell types that the following sections develop, each contributing a different piece of this representation. The first demonstration lets the reader see the frames come apart: a navigator who turns in place leaves every allocentric relation untouched while every egocentric bearing rotates.

Place Cells and the Cognitive Map

Tolman's map was an inference from behaviour; its neural reality was discovered by recording from single cells. In 1971 John O'Keefe and Jonathan Dostrovsky, recording from the hippocampus of freely moving rats, found neurons that fired only when the animal occupied a particular part of its environment and fell silent elsewhere. Each such place cell has a place field — a circumscribed region of space in which it is active — and different cells have fields in different places, so that at any moment the animal's location is encoded by which subset of the population is firing (O'Keefe & Dostrovsky, 1971). This was the first direct evidence that the brain contains an explicit representation of location, and it identified the hippocampus, already implicated in memory, as its seat.

O'Keefe and Lynn Nadel drew the theoretical conclusion in a book that gave the field its charter: the hippocampus, they argued, is the neural substrate of Tolman's cognitive map, an allocentric representation of space within which experiences are located and from which flexible navigation is computed (O'Keefe & Nadel, 1978). Place fields have properties that fit the role. They are allocentric — a field stays fixed to a place in the room as the animal approaches it from different directions — and they are shaped by the geometry of the enclosure and its landmarks, remapping to a new configuration of fields when the environment changes enough. The experimental workhorse for studying the behaviour that depends on this system is the Morris water maze, in which a rodent must find a platform hidden just beneath the surface of opaque water using only the distal landmarks around the room; because the platform gives no local cue, escape requires an allocentric fix on its location, and performance in the task is a standard assay of hippocampal spatial learning (Morris, 1984).

That the geometry of a space has a privileged role in fixing an allocentric representation was shown behaviourally before it was traced to any cell. Ken Cheng found that a rat disoriented in a rectangular chamber searches for a remembered goal not only at the correct corner but equally often at the diagonally opposite one — the corner that a rectangle's shape cannot distinguish from it — even when a distinctive featural cue, such as a differently coloured wall, is available to break the tie. The animal reorients by the geometry of the enclosure and largely ignores the feature, which led Cheng to propose an encapsulated geometric module that computes location from the shape of the environment alone (Cheng, 1986). The result launched a large comparative literature on reorientation across species and development, and it connects directly to the boundary-sensitive cells described below, which give the geometric module a plausible neural substrate in neurons tuned to the metric layout of environmental edges.

The hippocampal map is not the brain's only way of getting around. O'Keefe and Nadel themselves paired the flexible locale system, which reads the cognitive map, with a taxon system that learns fixed stimulus-response routes — turn right at the red barn — with no map at all. The two are dissociable in the brain: Mark Packard and James McGaugh showed that temporarily inactivating the rat hippocampus impairs place learning while sparing response learning, whereas inactivating the caudate nucleus of the dorsal striatum does the reverse, so the flexible map-based system and the habitual route-based system rest on separate neural substrates (Packard & McGaugh, 1996). Normal wayfinding blends the two, and which one governs behaviour shifts with practice and with the demands of the task — a distinction that returns below, because aging and early Alzheimer's disease push navigators off the flexible map and onto rigid, habitual routes.

Grid Cells and a Metric for Space

Place cells say where the animal is, but not, on their own, how far it is from anywhere else; a map needs a metric. That metric was found one synapse upstream. Recording in the medial entorhinal cortex, the main cortical input to the hippocampus, Marianne Fyhn, the Mosers, and their colleagues first saw cells with sharply localised fields like place cells but arranged with a striking regularity (Fyhn, Molden, Witter, Moser, & Moser, 2004). Mapping those fields over a larger enclosure revealed the pattern in full: each grid cell fires not in one location but in many, and those locations form a periodic triangular lattice tiling the whole environment, so that the firing fields sit at the vertices of tessellating equilateral triangles — a hexagonally symmetric grid (Hafting, Fyhn, Molden, Moser, & Moser, 2005). A grid cell is defined by three parameters: the spacing between fields, the orientation of the lattice relative to the environment, and its phase, or spatial offset. Because the lattice is regular and its geometry is intrinsic to the cell rather than tied to particular landmarks, the grid furnishes exactly the distance metric a place code lacks: displacement in any direction is read off as movement across a known, repeating scale. The discovery of grid cells, together with place cells, earned O'Keefe and the Mosers the 2014 Nobel Prize in Physiology or Medicine. The second demonstration lets the reader vary a grid's spacing and orientation and watch the lattice of firing fields respond.

A map with distances still needs a direction sense, and that is supplied by a separate population. James Taube and colleagues, recording in the postsubiculum, found head-direction cells, each of which fires whenever the animal's head points in a particular direction in the horizontal plane, regardless of where the animal is or what it is doing — a neural compass needle (Taube, Muller, & Ranck, 1990). Head-direction cells are anchored to the world by visual landmarks but persist in darkness, driven then by vestibular and self-motion signals, so heading is maintained even with the eyes closed. The medial entorhinal cortex also contains conjunctive cells that combine grid and head-direction tuning, while boundary vector cells — predicted by a computational model and then recorded in the subiculum — fire at a fixed distance and direction from an environmental border, tying the metric grid to the physical edges of the world (Lever, Burton, Jeewajee, O'Keefe, & Burgess, 2009). Together these cell types — place, grid, head-direction, and boundary — compose the allocentric representation, and the historical arc from Tolman's inference to this catalogue of tuned neurons has been told as one of cognitive neuroscience's cleanest success stories (Moser, Moser, & McNaughton, 2017). Table 1 summarises the four spatial cell types, the structure each was first recorded in, and the piece of the cognitive map each supplies.

Path Integration

The one navigational computation that needs no landmarks at all is path integration, also called dead reckoning. As the animal moves, it can update an estimate of its position by continuously summing its own velocity — direction and speed — over time, so that at every moment it holds a homing vector, the straight-line distance and bearing back to its start. A foraging ant that wanders a long crooked outbound path and then runs straight home the instant it finds food is performing path integration: it has integrated its self-motion into a single return vector, with no memory of the outbound route required. The computation draws on exactly the internal signals the previous section described — head-direction cells for the bearing of each step and a distance signal, to which the entorhinal grid is thought to contribute, for its length — which is why path integration is often taken as a core function of the grid-cell system.

Its virtue is also its weakness. Because path integration sums estimates, it sums their errors too: every imperfect reading of heading or distance adds to a drift that grows without bound the longer the animal relies on self-motion alone. A path-integration system left to itself therefore becomes steadily less accurate, which is why real navigators periodically reset it against stable external landmarks, correcting the accumulated drift whenever a recognised place comes into view. Navigation in practice is thus a continual interplay between two sources of position information — idiothetic self-motion cues that are always available but drift, and allothetic landmark cues that are stable but intermittent — each compensating for the other's failure mode. The third demonstration lets the reader lay out an outbound path and read off the homing vector the animal must compute to return, the same calculation worked by hand below.

The Human Navigation Network

The same architecture operates in people. The human medial temporal lobe contains place-like and grid-like responses, and neuroimaging has mapped a distributed network — hippocampus, entorhinal and parahippocampal cortex, retrosplenial cortex — that supports human wayfinding (Epstein et al., 2017). Two findings anchor the human case. First, Russell Epstein and Nancy Kanwisher identified a region of parahippocampal cortex, the parahippocampal place area, that responds far more strongly to images of places and scenes — rooms, landscapes, streetscapes — than to objects or faces, and responds especially to the geometry of the local spatial layout (Epstein & Kanwisher, 1998). It provides the visual scene analysis on which landmark-based navigation depends. Second, Eleanor Maguire and colleagues showed that navigational experience reshapes the human hippocampus itself: London licensed taxi drivers, who must memorise the city's tangle of streets to earn their licence, have significantly more grey matter in the posterior hippocampus than controls, and the volume scales with the number of years spent driving (Maguire et al., 2000). The result is a landmark demonstration that spatial expertise is written into brain structure.

The distinctively human question is whether the rodent grid code exists in people too, and Christian Doeller, Caswell Barry, and Neil Burgess supplied a signature that it does. Grid cells predict a specific fMRI consequence: because a population of grid cells shares a common lattice orientation, movement aligned with the grid axes should drive the entorhinal population differently from movement off-axis, producing a signal with six-fold rotational symmetry as a person's virtual heading changes. Doeller and colleagues found exactly this hexadirectional signal in the human entorhinal cortex during virtual navigation — indirect but compelling evidence for grid-like coding in the human brain (Doeller, Barry, & Burgess, 2010). Alongside these mechanisms sits wide variation between individuals: people differ markedly and stably in navigational ability, and that variation reflects a mixture of the strategies they adopt, their reliance on landmarks versus self-motion, their spatial anxiety, and experience (Wolbers & Hegarty, 2010).

When Navigation Fails: Aging and Alzheimer's

Because the entorhinal cortex and hippocampus are the seat of the navigation system, and because those same structures are among the first the pathology of Alzheimer's disease attacks, spatial navigation has a special clinical status: it is among the earliest cognitive functions to decline, often before standard memory tests detect impairment. Getting lost in familiar places is a common early symptom, and controlled tests of path integration and allocentric map use reveal deficits in people at genetic or biomarker risk while their episodic memory still appears intact. This has raised the prospect of navigation tasks as a sensitive, early cognitive marker for preclinical Alzheimer's disease, potentially discriminating incipient disease from healthy aging earlier than conventional instruments (Coughlan, Laczo, Hort, Minihane, & Hornberger, 2018). Aging itself degrades the grid-cell metric and shifts navigators away from flexible allocentric strategies toward more rigid route-following, so distinguishing normal age-related change from the early signature of disease is an active goal.

Beyond Physical Space

The most expansive claim in the field is that the cognitive map is not for physical space alone. The hippocampal-entorhinal system, on this view, implements a general-purpose format for representing the relational structure of any domain, and physical navigation is merely its most conspicuous use. Grid-like six-fold signals have been reported as people move through abstract, non-spatial feature spaces, as though the same metric machinery were mapping conceptual distances the way it maps metric ones, and the map has been proposed as the substrate of a spatial code for human thinking (Bellmund, Gardenfors, Moser, & Doeller, 2018). A refinement of the idea distinguishes two formats the system can use: a metric cognitive map, which supports Euclidean inferences like shortcuts and straight-line distances, and a more flexible cognitive graph, which represents places or concepts as nodes linked by transitions without a global metric, better capturing how people often navigate real, cluttered environments and knowledge alike (Peer, Brunec, Newcombe, & Epstein, 2021). On this account the study of spatial navigation reaches past wayfinding into the general question of how the brain represents structured knowledge.

Worked Example

Path integration can be worked by hand, and the third demonstration reproduces the calculation. Suppose an animal leaves its nest and walks three straight legs before stopping. It goes 8 metres east, then 12 metres north, then 3 metres west. To find its way home it must compute the single homing vector — the distance and direction back to the nest — from these movements alone.

The method is vector addition in a world-anchored (allocentric) frame. Take east as the positive x-axis and north as the positive y-axis, and add the legs component by component. The east-west total is 8 - 3 = 5 metres east; the north-south total is 12 metres north. So the animal's displacement from the nest is 5 metres east and 12 metres north. The straight-line distance home is the length of that displacement, √(5² + 12²) = √(25 + 144) = √169 = 13 metres — a clean 5-12-13 right triangle. The homing vector points from the animal's current position back to the nest, in the direction exactly opposite to the displacement. The displacement lies at a bearing of arctan(5 / 12) ≈ 22.6° east of north, so the animal must turn to face the reverse bearing, about 202.6° (that is, roughly south-southwest), and travel 13 metres to arrive home. Notice that the outbound path covered 8 + 12 + 3 = 23 metres, but the computed return is only 13 metres: path integration has thrown away the crooked route and kept only the net displacement, which is the entire point of holding a homing vector rather than a memory of the path. The estimate is only as good as the self-motion signals behind it; had each leg been misjudged by a few percent, those errors would compound in the 13-metre result, which is why an animal resets the computation against a landmark whenever it can.

Discussion

Spatial navigation is the case in which cognitive psychology's central strategy — inferring an unobservable internal representation from structured behaviour — was most fully vindicated by later physiology. Tolman's cognitive map was a behavioural inference that many at the time found extravagant; within three decades single-unit recording had found the map's cells, and within six it had catalogued a coordinated system of place, grid, head-direction, and boundary codes whose properties match the representational demands the behaviour implied (Tolman, 1948; O'Keefe & Nadel, 1978; Moser et al., 2017). The field's organising distinction, between an egocentric frame for action and an allocentric frame for flexible planning, remains the axis along which its mechanisms are understood, and path integration and landmark correction remain the two complementary sources of position that every navigator must reconcile.

Two open themes carry the subject forward. The first is the reach of the code: the proposal that the hippocampal-entorhinal map is a general engine for relational knowledge, applied to conceptual as readily as to physical space, would make navigation a model system for cognition at large rather than a specialised faculty, and the metric-map-versus-graph debate is where that proposal is now being sharpened (Bellmund et al., 2018; Peer et al., 2021). The second is clinical: because the navigation system's anatomy coincides with the earliest targets of Alzheimer's pathology, a well-chosen navigation task may detect the disease before conventional memory tests can, turning a basic-science story about rodent grid cells into a tool for early human diagnosis (Coughlan et al., 2018). That a line of work beginning with rats in a maze now bears on both the nature of human thought and the detection of dementia is a measure of how deep the problem of finding one's way turns out to run.

Common Misconceptions

A cognitive map is a literal picture of the environment stored in the brain.

The cognitive map is a functional representation of spatial relations, not an image. It is distributed across populations of place, grid, and head-direction cells encoding location, distance, and heading, and it is often distorted, incomplete, or better described as a graph of connected places than as a scale drawing (O'Keefe & Nadel, 1978; Peer et al., 2021).

Navigation requires seeing landmarks.

Path integration lets an animal track its position and compute a homing vector from self-motion signals alone, with no landmarks in view, as a foraging ant does in returning straight to its nest. Landmarks are used to correct the drift that path integration accumulates, not to make navigation possible in the first place (Wolbers & Hegarty, 2010).

Place cells and grid cells are the same discovery under two names.

They are distinct cell types in different structures doing different jobs. A place cell fires in a single location in the hippocampus; a grid cell fires at many locations arranged in a hexagonal lattice in the entorhinal cortex, supplying the distance metric that a place code lacks. They were discovered decades apart (O'Keefe & Dostrovsky, 1971; Hafting et al., 2005).

Glossary

Allocentric frame.

A world-centred reference frame that describes locations by their relations to each other and to fixed landmarks, independent of the navigator's position; the format of a cognitive map.

Allothetic cue.

Position information derived from the external world, chiefly landmarks and other sensory features of the environment; stable but available only intermittently.

Boundary vector cell.

A neuron that fires when an environmental boundary lies at a particular distance and direction from the animal, tying the metric map to the physical edges of the world.

Cognitive graph.

A representation of places or concepts as nodes linked by transitions, without a global metric; proposed as a more flexible alternative to a Euclidean cognitive map for cluttered environments and abstract knowledge.

Cognitive map.

An internal, allocentric representation of the spatial layout of an environment that supports flexible route-planning, shortcuts, and detours; inferred by Tolman and located by O'Keefe and Nadel in the hippocampus.

Egocentric frame.

A body-centred reference frame that describes locations by their bearing and distance relative to the navigator; immediate and needed for action, but invalidated whenever the body moves.

Entorhinal cortex.

The main cortical gateway to the hippocampus and the site of grid cells, head-direction cells, and boundary cells; source of the metric signals the hippocampal map draws on.

Geometric module.

A hypothesised encapsulated system, proposed by Cheng, that computes an animal's location and heading from the metric shape of the enclosing space, dominating reorientation even when distinctive featural landmarks are available to disambiguate it.

Grid cell.

An entorhinal neuron that fires at many locations arranged in a regular hexagonal lattice tiling the environment, providing a distance metric; characterised by its spacing, orientation, and phase.

Head-direction cell.

A neuron that fires whenever the animal's head points in a particular horizontal direction regardless of location, functioning as a neural compass anchored by landmarks but sustained by self-motion in darkness.

Hippocampus.

A medial temporal lobe structure containing place cells and, in O'Keefe and Nadel's theory, the neural substrate of the cognitive map; also central to episodic memory.

Homing vector.

The running estimate of the straight-line distance and direction from the navigator's current position back to its starting point, maintained by path integration.

Idiothetic cue.

Position information derived from the animal's own movement — vestibular, proprioceptive, and motor-efference signals; always available but subject to accumulating drift.

Parahippocampal place area.

A scene-selective region of parahippocampal cortex that responds strongly to places and to the geometry of the local spatial layout, supporting landmark-based navigation.

Path integration.

Also dead reckoning; the continuous updating of a position estimate by summing self-motion over time to maintain a homing vector, requiring no landmarks but accumulating error.

Place cell.

A hippocampal neuron that fires selectively when the animal occupies a particular region of the environment — its place field — so that the active population encodes current location.

Reference frame.

The origin and axes with respect to which positions are represented; navigation uses egocentric (body-centred) and allocentric (world-centred) frames and continually translates between them.

Response learning.

Habitual, map-free navigation in which a fixed stimulus-response route is learned (the taxon system); dissociable from hippocampal place learning and dependent on the dorsal striatum.

Spatial navigation.

The mental process of determining and following a route through space by tracking one's own position and heading and computing a path to a goal.

Survey knowledge.

An integrated, map-like allocentric representation of how the places in an environment lie relative to one another; in Siegel and White's account, the final stage acquired after landmark and route knowledge, supporting shortcuts and novel routes.

Key Researchers

Neil Burgess. Professor of Cognitive and Computational Neuroscience at University College London; he built computational models linking place, grid, and boundary-vector cells to human spatial memory and provided fMRI evidence for grid-like coding in the human entorhinal cortex. Faculty Page - Google Scholar - Wikipedia - ORCID

Russell A. Epstein. Professor of Psychology at the University of Pennsylvania; he discovered the parahippocampal place area, a scene-selective visual region, and maps how the human brain represents environments and headings during real-world navigation. Faculty Page - Google Scholar - Wikipedia

Eleanor A. Maguire (1970-2025). Late Professor of Cognitive Neuroscience at University College London; her structural MRI studies of London taxi drivers showed that navigational expertise enlarges the posterior hippocampus, landmark human evidence that spatial experience reshapes the brain. Wikipedia - Wikidata - ORCID

Edvard I. Moser. Professor of Neuroscience and founding director of the Kavli Institute for Systems Neuroscience at the Norwegian University of Science and Technology; he co-discovered grid cells and mapped the entorhinal circuitry supporting path integration, sharing the 2014 Nobel Prize. Faculty Page - Google Scholar - Wikipedia - ORCID

May-Britt Moser. Professor of Neuroscience at the Norwegian University of Science and Technology and director of its Centre for Neural Computation; she co-discovered grid cells in the entorhinal cortex, whose periodic firing provides a metric coordinate system for space, sharing the 2014 Nobel Prize. Faculty Page - Google Scholar - Wikipedia - ORCID

John O'Keefe. Emeritus Professor at University College London and director of the Sainsbury Wellcome Centre; he discovered place cells in the rat hippocampus and, with Nadel, developed the cognitive-map theory of the hippocampus, for which he received the 2014 Nobel Prize. Faculty Page - Wikipedia - ORCID

Edward C. Tolman (1886-1959). Psychologist at the University of California, Berkeley; from latent-learning experiments he argued that rats build an internal cognitive map of a maze rather than a chain of stimulus-response habits, introducing the founding idea of spatial cognition. Wikipedia - Wikidata

Frequently Asked Questions

What is spatial navigation in cognitive psychology?
It is the mental process of determining and following a route through space: tracking one's own position and heading, representing where places and goals are, and computing a path to a goal that may be out of sight. Because the needed information is rarely in the immediate stimulus, it depends on an internal representation of space (Epstein et al., 2017).

What is a cognitive map?
A cognitive map is an internal, world-centred (allocentric) representation of the spatial relations in an environment that lets an animal plan routes, take shortcuts, and make detours it has never practised. Tolman inferred it from rats' latent learning, and O'Keefe and Nadel argued it is implemented by the hippocampus (Tolman, 1948; O'Keefe & Nadel, 1978).

What is the difference between egocentric and allocentric navigation?
Egocentric representations describe locations relative to the navigator's own body (ahead, to the left) and are needed for action but change whenever the body moves. Allocentric representations describe locations relative to the world and fixed landmarks, stay stable as the navigator moves, and support flexible route-planning; the cognitive map is allocentric (Burgess et al., 2002).

What are place cells and grid cells?
Place cells are hippocampal neurons that fire when the animal is in a particular location, so the active population signals where it is. Grid cells are entorhinal neurons that fire at many locations arranged in a hexagonal lattice, providing a distance metric. Their discovery earned the 2014 Nobel Prize (O'Keefe & Dostrovsky, 1971; Hafting et al., 2005).

What is path integration?
Path integration, or dead reckoning, is keeping a running estimate of position by summing an animal's own movements over time, so it always holds a homing vector, the straight-line distance and direction back to its start. It needs no landmarks but accumulates error, so it is periodically reset against them (Wolbers & Hegarty, 2010).

Do London taxi drivers really have bigger hippocampi?
Yes. Maguire and colleagues found that London licensed taxi drivers, who memorise the city's street layout, have more grey matter in the posterior hippocampus than controls, with the volume increasing with years of driving, evidence that navigational experience reshapes brain structure (Maguire et al., 2000).

Why is spatial navigation relevant to Alzheimer's disease?
The entorhinal cortex and hippocampus that house the navigation system are among the first regions Alzheimer's pathology attacks, so navigation often declines very early, before standard memory tests detect impairment. This makes navigation tasks a candidate early cognitive marker for preclinical Alzheimer's disease (Coughlan et al., 2018).

Is the cognitive map used only for physical space?
Probably not. Grid-like signals appear as people move through abstract, non-spatial feature spaces, suggesting the hippocampal-entorhinal system provides a general format for relational knowledge, of which physical navigation is one use; whether it is best described as a metric map or a more flexible graph is debated (Bellmund et al., 2018; Peer et al., 2021).

References