Spatial Navigation

Abstract

Spatial navigation, which MeSH classifies under both spatial behavior and mental processes, is the use of cues to plan a route to a destination and then to follow it. Research on the topic centres on map-like representations of space and on the place, grid and head-direction cells of the hippocampal-entorhinal system that are thought to support them. A second line of work concerns path integration from self-motion and the contrast between flexible place strategies and habitual response strategies. Studies of people add brain imaging, experience-related change in the hippocampus, large individual differences, and the uneven decline of navigation with age and in early Alzheimer's disease. Three interactive demonstrations accompany the article: a path-integrating walker, an adjustable grid-cell firing map, and a place-versus-response maze test.

Keywords: spatial navigation, cognitive map, place cells, grid cells, path integration

How animals and people reach a goal they cannot see is a question that ties behaviour, cognition and the firing of single neurons closely together. The best-studied part of the answer lies in the hippocampus and the neighbouring entorhinal cortex. A review of three decades of recording there concluded that hippocampal place cells and entorhinal grid cells jointly represent where an animal is, while treating the grid cells' contribution to path integration as tentative (Moser et al., 2008).

Human research has since moved in the same direction. Taking the rodent findings as its starting point, a 2017 review argued that the human hippocampus and entorhinal cortex also carry map-like codes for space, and it discussed how those codes might be applied to nonspatial domains (Epstein et al., 2017). This article follows navigation from the idea of an internal map, through the cells and strategies that implement wayfinding, to its variation across people, its change with experience, and its decline with age and disease.

What Spatial Navigation Is

Process by which organisms use various cues to develop a route to a destination and their ability to travel such a planned route.

Scope note for the MeSH descriptor Spatial Navigation (D065854), U.S. National Library of Medicine.

MeSH lists two entry terms under this descriptor, Spatial Ability and Spatial Visualization, which tie navigation to the broader capacity to represent and transform spatial information.

The definition has two halves, planning a route and executing it, and the descriptor sits accordingly in two places in the MeSH tree: under spatial behavior, as something organisms do, and under mental processes, as something they compute. Spatial ability, the first entry term, refers to the general capacity to encode, remember and reason about spatial relations; spatial visualization, the second, refers more narrowly to imagining spatial forms and transforming them mentally, as in rotating or folding an object in the mind. Neither is identical to wayfinding through a real environment, and that gap is the territory of a self-report measure developed specifically for environmental spatial ability (Hegarty et al., 2002).

Any navigator must express locations in some frame of reference. In an egocentric frame, positions are coded relative to the navigator's own body (ahead, to the left, behind), so the code changes with every step and turn. In an allocentric frame, positions are coded relative to the environment or to one another, so the code stays fixed as the navigator moves. How far animals and people rely on each is disputed. Wang and Spelke (2002) argued from comparisons of human and animal navigation that egocentric representations updated from moment to moment do much of the work usually credited to enduring geocentric maps. They presented this as a challenge to the widely held view, and in their account human navigation builds on systems shared with other animals.

People also differ widely in how well they navigate. A review of those differences located their sources in three interdependent domains: variation in brain microstructure, variation in how neural circuits process spatial information, and cognitive and perceptual factors; the authors offered this picture as provisional and still emerging (Wolbers & Hegarty, 2010). Navigation overlaps with several neighbouring constructs: spatial memory, the retention of locations and layouts; spatial orientation, knowing where one is and which way one faces; and space perception, the perceptual registration of distance, direction and layout on which both depend.

Cognitive Maps

Cognitive map theory holds that animals form an internal, map-like representation of their environment and use it to plan new routes, and it is the organising idea of the field. Its behavioural origin is Tolman's paper on cognitive maps in rats and men (Tolman, 1948); its neural form came three decades later in a book proposing that the hippocampus is the brain's cognitive map (O'Keefe & Nadel, 1978). The theory has since been revised rather than discarded. A review of human work found map-like codes in the hippocampus and entorhinal cortex that fit the core idea (Epstein et al., 2017). Stachenfeld et al. (2017) used a reinforcement learning model to recast the hippocampus as a predictive map, one that represents what is likely to come next and so is not purely spatial; they argued, without showing, that entorhinal grid cells play a part in it. Behrens et al. (2018) went further, treating spatial maps as one instance of a general code for organising abstract, generalisable knowledge in the service of flexible behaviour, while noting that the neural evidence remains mostly spatial, that part of their support comes from artificial agents, and that their final claims are speculative. A long-standing rival, relational theory, holds that the hippocampus processes relations of any kind, not space in particular. When an fMRI study set a spatial task against a matched task about social networks, hippocampal activity appeared biased toward spatial relations, a result its authors took as favouring cognitive map theory (Kumaran & Maguire, 2005). Earlier, Wang and Spelke (2002) had challenged the premise that navigation needs an enduring map at all, putting momentary egocentric representations in its place. A more recent review contrasts Euclidean cognitive maps with cognitive graphs, in which places are joined by paths, and argues that both kinds of representation exist and rest on partly overlapping neural systems (Peer et al., 2021). The theory therefore survives in revised form, and whether one stable map underlies wayfinding remains unresolved.

The broadest version of the revision concerns thought rather than travel. Bellmund et al. (2018) proposed that a geometric code in the hippocampal-entorhinal region organises cognitive spaces in high-level cognition generally, so that conceptual knowledge would be traversed much as a room is; the authors discussed recent evidence for the proposal.

Place Cells and Replay

The neural search for the cognitive map began with recordings of single units in the hippocampus. O'Keefe and Dostrovsky's short report offered preliminary evidence, from unit activity in freely moving rats, that the hippocampus acts as a spatial map (O'Keefe & Dostrovsky, 1971). A decade later a lesion study reported, as its title states, that rats with hippocampal lesions were impaired at place navigation (Morris et al., 1982), tying the structure to the behaviour that the map idea predicts.

Place cells do more than mark the animal's present position. Recording in rats, Foster and Wilson (2006) found that sequences of place-cell activity were replayed while the animals were awake, in the reverse of their original temporal order. Linking the result to reinforcement learning models, they proposed that such replay could support spatial learning, a role their data did not test directly. Diba and Buzsáki (2007) recorded rats on an elevated track and found sequences in both directions, timed with sharp waves: reversed sequences after a run and forward sequences before one. They raised a possible role in episodic memory without establishing it; that link connects replay to wider questions about memory consolidation that lie outside this article.

Evidence that replay can look ahead comes from an open arena. Pfeiffer and Foster (2013) recorded with 40 tetrodes from 4 rats, each studied on two days, in a 2m x 2m arena with 36 possible goal locations. Bayesian decoding identified 2028 trajectory events, and these sequences tended to trace the route the rat was about to take toward a goal it had remembered; the rats were not biased to face that goal beforehand, so the effect was not a matter of heading. The sample was only four animals, and the authors left open whether such events would compute good routes in tasks with barriers to movement. Jadhav et al. (2012) interrupted sharp-wave ripples, the brief events during which replay occurs, in awake rats on a W-shaped track, and performance fell in a way the authors inferred to be a deficit of spatial working memory, while place fields stayed intact. The groups of rats were small, however, and the method disrupted the ripple events as a whole rather than the content of replay in particular.

Grid and Head-Direction Cells

A map needs a compass. Taube et al. (1990) described 24 head-direction cells from 7 rats recorded in the postsubiculum while the animals moved freely in a cylindrical chamber marked by one white-card cue, their heading tracked by video from 2 colored LEDs. Each cell fired most strongly when the head pointed in its preferred direction, with a roughly triangular firing function, and the cells' peak firing rates spanned 5 to 115 spikes/sec, with a mean of 35. The authors classified these cells as firing independently of the animal's location, and quantitative analysis showed that location had a minimal effect on their firing.

The grid cell supplied a metric. Hafting et al. (2005) recorded single units in dorsocaudal medial entorhinal cortex and found cells that each fired at multiple locations forming an equilateral triangular grid across the environment, with cells differing in the phase of their grids. The grid persisted when landmarks were removed, which led the authors to suggest, without demonstrating, that the cells take part in path integration. In their review, Moser et al. (2008) placed grid cells and place cells in one spatial representation system for self-location, with the grid input stated as something that may support path integration rather than as an established mechanism.

The demonstration below draws a single grid cell's firing map as a triangular lattice beside a single place field. Its sliders change the lattice's spacing and orientation; the display is illustrative and contains no recorded data.

Path Integration

Path integration is the computation of current position from self-motion alone, by continuously adding up the direction and distance of each movement since leaving a known point. Reviewing work in mammals, Etienne and Jeffery (2004) described how path integration keeps an animal oriented when no landmarks are available, how it interacts with location-based cues when they are, and how it feeds the internal representation of space; they gave particular weight to connecting the neural evidence with behaviour.

The discovery of grid cells offered a candidate substrate. McNaughton et al. (2006) proposed, combining earlier theoretical studies with then-recent empirical findings, that the medial entorhinal cortex might perform some of the essential computations of path integration and that convergent input from grid cells at multiple scales might give rise to place fields, with spatial scale increasing along the dorsoventral axis. Gil et al. (2018) later disrupted grid-cell firing in mice through a genetic manipulation that left other spatial cells unaffected, and the animals' path integration was impaired. Because that evidence comes from mice only, how far grid cells carry out path integration in other species, and what else they compute, remains open (see Discussion).

In the demonstration below, a simulated walker takes a random outbound path, keeps a running estimate of its position from noisy self-motion signals, and then heads for home. The gap between where it believes home lies and where home actually is widens as the path lengthens and the noise grows; the simulation is illustrative, not data, and uses the same arithmetic as the Worked Example.

Navigation Strategies

A navigator can reach the same goal in two quite different ways: by heading for a location defined by its relation to surrounding cues, or by repeating a sequence of turns that has worked before. Packard and McGaugh (1996) separated the two in rats trained on a cross-maze with four trials a day and only extramaze visual cues. On probe trials given on Day 8 and Day 16, a single probe per day, they temporarily inactivated either the hippocampus or the caudate nucleus with 2% lidocaine. Hippocampal inactivation interfered with place learning, whereas caudate inactivation interfered with response learning, a double dissociation central to the study of maze learning. The caudate nucleus belongs to the basal ganglia, and response strategies are accordingly often described as the spatial counterpart of habits.

The same division appears in people, with a twist: individuals choose. In an fMRI study using a computer-generated virtual place-learning task, Iaria et al. (2003) found that participants split about half and half between a spatial strategy based on landmarks and a nonspatial one. Landmark users engaged the right hippocampus and nonspatial users the caudate nucleus, both groups shared activation in posterior parietal and frontal cortex, and strategies shifted as practice continued.

Everyday technology may push people toward one route. In a study of 50 regular drivers, Dahmani and Bohbot (2020) found that greater lifetime GPS experience went with poorer spatial memory during self-guided navigation, measured with tasks including the 4-on-8 Virtual Maze, and they retested 13 of the participants three years later. The evidence is correlational, and the authors only suggested a causal reading. The three-year follow-up was small and unplanned, the cross-sectional correlation with a combined probe measure (r = −0.20) was only marginally significant, and the authors called for replication in a larger sample.

The demonstration below shows the classic test that separates the two strategies. An animal trained to find food from one start arm of a plus-shaped maze is released from the opposite arm: a place strategy sends it to the same location, while a response strategy repeats the same turn and sends it the wrong way. The display is illustrative, not data.

Navigation in the Human Brain

Direct recordings from the human brain are rare, and the clearest come from patients. Ekstrom et al. (2003) recorded 317 neurons in patients as they navigated a virtual town and found place cells in the hippocampus, spatial view cells in the parahippocampal region, and goal cells. The study showed that cell types like those described in rats exist in people, though the evidence comes from a clinical sample.

Combining rat electrophysiology with fMRI in a virtual reality environment, Doeller et al. (2010) found a signal that varied with running direction in a pattern of six-fold rotational symmetry, the signature expected from grid cells, across an entorhinal network tied to spatial and autobiographical memory. The signal was strongest in the right entorhinal cortex, and it is indirect fMRI evidence, not a recording of single cells. Jacobs et al. (2013) later recorded single neurons with grid-like firing directly, among 893 cells from 14 epilepsy patients who performed a virtual navigation task. Those firing maps were noisier than those of some rodent grid cells, and the virtual task gave no proprioceptive feedback, which makes extrapolation to real-world movement difficult.

Navigation experience also appears to reshape the hippocampus. Using structural MRI, Maguire et al. (2000) found that licensed London taxi drivers had larger posterior hippocampi than controls, smaller anterior hippocampi, and posterior volumes that rose with time spent as a driver. A correlational group comparison of this kind cannot say whether the job enlarged the structure or people with larger posterior hippocampi became drivers.

Woollett and Maguire (2011) addressed that question longitudinally. They scanned 79 male trainee drivers and 31 male controls before and after the 3–4 years that trainees spend learning the Knowledge, the layout of about 25,000 streets within a 6-mile radius. Of the trainees retested, 39 had qualified and 20 had not; posterior hippocampal grey matter increased only in those who qualified, not in those who did not or in controls, and no anterior decrease was seen. The qualifiers were, however, poorer at recalling the Taylor complex figure, a test of visual memory; whether gains in spatial knowledge carry costs for other kinds of memory, including episodic memory, is not settled by this result. The authors could not exclude a genetic predisposition among successful trainees, MRI could not identify the cellular mechanism, and 20 of the 79 trainees were not rescanned.

Individual Differences and Ageing

Measuring navigation ability is the first problem in studying how people differ. Hegarty et al. (2002) developed a self-report measure aimed specifically at environmental spatial ability. The review by Wolbers and Hegarty (2010) described how differences in brain structure, in neural processing and in cognitive and perceptual factors interact to produce good and poor navigators, while presenting that picture as unfinished.

The largest data set comes from a game. Coutrot et al. (2018) assessed navigation in more than 2.5 million people across many countries with a mobile-app-based cognitive task. Navigation ability clustered into five groups of countries: a nation's economic wealth predicted the average ability of its inhabitants, and gender inequality predicted the size of the performance difference between men and women. These associations predict performance but do not establish what causes the differences.

Age brings a decline that is not uniform. Drawing on studies of rodents, non-human primates and humans, Lester et al. (2017) concluded that ageing impairs navigation unevenly across its component computations and that the deficits are not just one more sign of general memory decline. They pointed to navigation measures as potential markers of pathology with clinical promise, a theme that connects this work to cognitive aging research more broadly.

Alzheimer's disease is the main pathology in question. Kunz et al. (2015) compared people at risk of developing Alzheimer's disease with controls during virtual navigation and reported that the at-risk group differed in its entorhinal signal, many decades before any expected onset, and also moved through the virtual environment differently. The available summary of that study is an editor's précis and gives no sample size, so the size of the effect cannot be judged here. In a narrative review, Coughlan et al. (2018) argued that navigation deficits are an overlooked cognitive marker of preclinical Alzheimer's disease, noting that standard episodic memory tests have low specificity and that evidence from at-risk individuals is accumulating; they also acknowledged that research gaps remain.

Discussion

Several conclusions rest on converging evidence. Place cells, grid cells and head-direction cells have been recorded repeatedly in rodents, cells of the place type and cells with grid-like firing have been recorded directly in people, and grid-like signals have also been inferred from human imaging. The separation between hippocampus-dependent place strategies and caudate-dependent response strategies holds in rat inactivation studies and in human imaging, and the human work adds that people differ in which strategy they adopt and change it with practice. The hippocampus also changes structurally with intensive navigational learning, at least in the taxi trainees who went on to qualify.

Other questions remain open, and they share a common thread: how far the clean cellular picture explains behaviour. The first is whether navigation rests on a single enduring allocentric map at all, as cognitive map theory in its original form held, or on momentary egocentric representations as Wang and Spelke argued; the revised forms of the theory have not settled this. A graph of connected places is one current alternative to a metric map (Peer et al., 2021). The second is what grid cells actually compute. Disrupting grid cells in mice impaired path integration (Gil et al., 2018), but whether that is their main job, or whether they also serve the predictive role others propose, is unsettled. The third is how far map-like codes extend beyond physical space; the most ambitious of the reviews that make that case calls its broadest claims speculative and notes that the neural evidence is still mostly spatial. The fourth is what causes individual differences. The taxi studies cannot rule out predisposition, the GPS study is correlational with a small follow-up, and the global app data are associations. Settling these questions would take designs that manipulate experience in advance, such as randomised training or restricted GPS use, together with manipulations of grid-cell activity beyond the single mouse study.

Current Directions

The most active theoretical front reinterprets the cognitive map. The predictive-map proposal (Stachenfeld et al., 2017), the treatment of maps as a general scheme for structuring knowledge (Behrens et al., 2018) and the proposal that a hippocampal-entorhinal geometric code organises conceptual spaces (Bellmund et al., 2018) all extend navigation research toward reasoning and planning in general (see Cognitive Maps). Human work increasingly asks whether the map-like codes found in rodents operate the same way in people (Epstein et al., 2017).

A second front is population-scale and clinical. Mobile games now measure navigation in millions of people and relate it to national conditions such as wealth and gender inequality, at the level of association (Coutrot et al., 2018). Work on ageing is separating which navigational computations decline and arguing that the pattern is not reducible to general memory loss (Lester et al., 2017), and navigation tasks are being examined as candidate early markers of Alzheimer's disease (Coughlan et al., 2018). Finally, the effect of everyday navigation aids on spatial memory has become a research question in its own right, so far with correlational evidence and a small longitudinal sample (Dahmani & Bohbot, 2020).

Common Misconceptions

The brain stores a single, fixed map of the world, like a paper map.

Whether any enduring allocentric map exists is disputed. Wang and Spelke argued that much navigation relies instead on egocentric representations updated moment to moment (Wang & Spelke, 2002), and later proposals recast the map as predictive or as a general knowledge code. The paper-map image survives because the term cognitive map suggests one.

Place cells alone explain how animals navigate.

Place cells are one element of a wider system. A major review placed them alongside entorhinal grid cells in a joint representation of self-location (Moser et al., 2008), and head-direction cells and path integration contribute further. The belief persists because place cells were discovered first and are the best known.

Some people simply have no sense of direction, and nothing can change it.

The hippocampus can change with navigational learning: trainee London taxi drivers who qualified after years of study showed growth in posterior hippocampal grey matter, while those who failed and controls did not (Woollett & Maguire, 2011). The authors could not exclude a genetic predisposition, so the study shows plasticity in those who succeeded rather than proving that anyone can become an expert navigator.

Using GPS has no effect on spatial memory.

Among regular drivers, heavier lifetime GPS use was associated with poorer spatial memory during self-guided navigation (Dahmani & Bohbot, 2020). The evidence is correlational and its longitudinal sample was small and unplanned, so it suggests rather than proves a cost; it does not support the claim of no effect.

Glossary

Allocentric reference frame.

A way of coding locations relative to the environment or to other objects, so that the code does not change as the observer moves.

Cognitive map theory.

The proposal that animals form an internal, map-like representation of their surroundings, later linked to the hippocampus; its standing is revised, with recent versions treating the map as predictive or as a general code for knowledge, and a rival view holding that navigation relies on momentary egocentric representations.

Egocentric reference frame.

A way of coding locations relative to the observer's own body, such as ahead or to the left, so that the code changes with every movement.

Entorhinal cortex.

A region of cortex adjacent to the hippocampus and closely connected with it; its medial part contains grid cells.

Grid cell.

A neuron in the medial entorhinal cortex that fires at many locations forming a regular triangular lattice across an environment.

Head-direction cell.

A neuron that fires when the head points in a particular direction, with only minimal dependence on where the animal is.

Hippocampus.

A structure in the medial temporal lobe that contains place cells and supports place-based navigation and memory.

Landmark.

A stable, recognisable feature of the environment used to fix position or direction and to correct errors in self-motion estimates.

Path integration.

Tracking one's position relative to a starting point by summing the direction and distance of one's own movements, without reference to landmarks.

Place cell.

A hippocampal neuron that fires mainly when the animal occupies one region of its environment, called the cell's place field.

Place learning.

Learning to reach a goal defined by its location relative to surrounding cues, a strategy that depends on the hippocampus.

Replay.

The rapid reactivation of a sequence of place cells, forwards or backwards, outside the movement that first produced it; interrupting the events that carry it has impaired spatial working memory in rats.

Response learning.

Learning to reach a goal by repeating a fixed movement, such as a particular turn, a habit-like strategy that depends on the caudate nucleus.

Spatial ability.

The general capacity to encode, remember and reason about spatial relations; a MeSH entry term under spatial navigation.

Spatial visualization.

The ability to imagine spatial forms and transform them mentally, for example by rotating or folding them; a MeSH entry term under spatial navigation.

Key Researchers

Véronique D. Bohbot

(McGill University; Douglas Mental Health University Institute). Neuroscientist who studies how people differ in using hippocampal, map-like strategies or caudate, habit-like strategies to navigate, and how GPS use relates to spatial memory. Faculty page

Neil Burgess

(University College London). Cognitive neuroscientist who studies spatial memory and found grid-cell-like signals in the human brain. ORCID · Wikipedia · Wikidata

Eleanor A. Maguire

(University College London; 1970–2025). Cognitive neuroscientist whose studies of London taxi drivers linked navigation experience to hippocampal structure. ORCID · Faculty page · Wikipedia · Wikidata

Edvard I. Moser

(Kavli Institute for Systems Neuroscience, NTNU). Neuroscientist who co-discovered grid cells in the entorhinal cortex. ORCID · Faculty page · Wikipedia · Wikidata

May-Britt Moser

(Kavli Institute for Systems Neuroscience, NTNU). Neuroscientist who co-discovered grid cells in the entorhinal cortex. ORCID · Faculty page · Wikipedia · Wikidata

John O'Keefe

(University College London). Neuroscientist who discovered hippocampal place cells and, with Lynn Nadel, proposed the hippocampus as a cognitive map. ORCID · Faculty page · Wikipedia · Wikidata

Hugo J. Spiers

(University College London). Cognitive neuroscientist who studies human navigation, including large-scale data on navigation ability across countries. ORCID

Edward C. Tolman

(University of California, Berkeley; 1886–1959). Psychologist who argued from rats' maze behaviour that animals learn cognitive maps of their surroundings. Wikipedia · Wikidata

Frequently Asked Questions

What is spatial navigation?

Spatial navigation is the use of cues to plan a route to a destination and then follow it. Much of the research concerns the hippocampus and entorhinal cortex, where place cells and grid cells are thought to represent an animal's own location (Moser et al., 2008).

What is a cognitive map?

A cognitive map is an internal, map-like representation of an environment that allows new routes to be planned. The idea has been revised: one influential proposal treats spatial maps as one case of a general code for organising knowledge, while conceding that its final claims are speculative (Behrens et al., 2018).

What do grid cells do?

Grid cells in the medial entorhinal cortex fire at many locations arranged in a triangular lattice across an environment, and the pattern persists without landmarks. Their discoverers suggested, without demonstrating, that they contribute to path integration (Hafting et al., 2005). A later mouse study that disrupted grid-cell firing found path integration impaired (Gil et al., 2018).

What is path integration?

Path integration is the tracking of position by summing the direction and distance of one's own movements. It allows an animal to stay oriented without landmarks, and in mammals it interacts with location-based cues when these are available (Etienne & Jeffery, 2004).

Do people use the same brain systems as rats to navigate?

Partly, on current evidence. Recordings in patients navigating a virtual town identified place cells in the human hippocampus, along with spatial view cells and goal cells (Ekstrom et al., 2003).

Why do some people navigate better than others?

Navigation ability varies with cognitive and perceptual factors, with how neural circuits process spatial information, and with differences in brain microstructure. A review of these sources presented the picture as still emerging (Wolbers & Hegarty, 2010).

Does navigation change with age?

Navigation declines with age, but unevenly across its component computations. A review of animal and human studies concluded that the decline is not simply part of general memory loss (Lester et al., 2017).

Can navigation problems signal Alzheimer's disease?

Possibly. A narrative review argued that navigation deficits are an overlooked marker of preclinical Alzheimer's disease, while acknowledging that research gaps remain (Coughlan et al., 2018).

References