Spatial Learning

Abstract

Spatial learning is a type of learning, and a form of spatial behavior: the acquisition of knowledge about the layout of the environment and the locations of objects, boundaries, and goals within it. It is what allows an organism to find a place again or take a novel shortcut, and it is isolated in tasks where the only thing to be learned is where something is. Two forms are distinguished: an allocentric form, coding locations relative to external landmarks in a viewpoint-independent map, and an egocentric form, coding them relative to the observer's own body. Converging lesion, single-unit, and imaging evidence identifies the hippocampus and medial temporal lobe as the core substrate of allocentric spatial learning, with place cells and entorhinal grid cells providing its positional code. This article covers its forms, neural basis, and measurement.

Keywords: spatial learning, cognitive map, place cells, hippocampus, allocentric

Spatial learning is among the most thoroughly analyzed forms of learning in psychology and neuroscience, in part because a location is an unusually clean thing to have learned: an animal either goes to the right place or it does not, and the route it takes reveals what it knows. The modern study of the topic begins with the observation that animals learn more than the movements that were rewarded. Rats trained to run a maze behave, when the maze is altered, as though they had acquired a map of its space rather than a chain of turns, a finding that led Tolman to propose the cognitive map as the internal representation guiding spatial behavior (Tolman, 1948). The subsequent discovery that single hippocampal neurons fire selectively when an animal occupies a particular location gave that abstract idea a physical substrate and made spatial learning a bridge between behavior and its neural mechanism (O'Keefe & Dostrovsky, 1971; O'Keefe & Nadel, 1978).

What Spatial Learning Is

Spatial learning is the process by which an organism acquires and stores information about the spatial structure of its environment: where landmarks, boundaries, resources, and goals are located, and how they are arranged with respect to one another. It is distinguished from other forms of learning by its content rather than its mechanism — what is learned is a set of spatial relations — and it is isolated experimentally by holding everything else constant so that only location carries the reward. An animal that learns to find a hidden platform in a pool, or to remember which arms of a maze it has already visited, has learned something spatial in exactly this sense (Morris, 1984; Olton & Samuelson, 1976).

The central theoretical distinction is between two frames of reference in which a location can be coded. In an egocentric frame, positions are represented relative to the observer — the goal is to the left, or two body-turns ahead — and learning consists of a route or a sequence of responses tied to the learner's own movements. In an allocentric frame, positions are represented relative to features of the world itself, independent of where the observer happens to be standing, and learning yields a maplike representation that supports novel shortcuts and flexible detours. The cognitive-map theory holds that the hippocampus builds the allocentric representation, while egocentric route learning can proceed through other systems (O'Keefe & Nadel, 1978). The two frames are not mutually exclusive; most natural navigation combines them, and the same destination can be reached either by recomputing position on a map or by replaying a learned sequence of turns.

That an animal can use a map at all — rather than only a chain of remembered responses — was Tolman's central claim, and the phenomena he marshaled for it, latent learning and place learning, remain the defining demonstrations that spatial learning produces knowledge of where, not merely a habit of how (Tolman, 1948). Spatial learning in this maplike sense is the acquisition process that yields spatial memory and supports spatial navigation; it is one component of the broader repertoire of spatial behavior.

Types of Spatial Learning

In the MeSH classification, spatial learning sits directly beneath learning and has a single narrower descriptor, maze learning, the tradition of tasks in which an animal learns a route or a location within a physical maze. The subtype below is a way the literature is indexed, not a claim that spatial learning decomposes into exactly one kind; the allocentric/egocentric distinction developed above cuts across it, and a maze-learning study may probe either frame. MeSH is an indexing vocabulary for the biomedical literature, so the list reflects how studies are catalogued rather than a theory of how spatial cognition is organized.

Table 1

Direct Subtypes of Spatial Learning in the MeSH Classification (tree F02.463.425.874)

Note. Maze learning (D018782) is the only narrower descriptor of spatial learning in the current MeSH tree; it is not yet a separate article and is shown here as plain text. Subtypes in an indexing vocabulary need not be mutually exclusive or jointly exhaustive.

Reference Frames and Learning Strategies

The distinction between allocentric and egocentric coding becomes an empirical question in the plus-maze, which pits a place strategy against a response strategy. A rat is trained to find food in one arm of a plus-shaped maze from a fixed start arm. Because the start is always the same, two things are confounded: the food is both in a particular place in the room and at the end of a particular body-turn. To separate them, the animal is started from the opposite arm on a probe trial. A learner using a place strategy — an allocentric map anchored to room cues — turns toward the same location in the room, which now requires the opposite body-turn. A learner using a response strategy — an egocentric habit — makes the same turn it was trained to make, arriving at the wrong place (Tolman, 1948). Which strategy dominates depends on training, on the availability of distal cues, and on which neural systems are intact.

This is why the availability of stable, distant landmarks is decisive for allocentric learning. When the environment offers reliable distal cues, animals and humans preferentially build a map; when it does not, they fall back on egocentric routes. The two systems can also compete, and damage to the hippocampal system shifts behavior toward striatally mediated response learning, so that an animal unable to build a map can still acquire the task as a fixed sequence of turns (O'Keefe & Nadel, 1978). The demonstration below lets the same start displacement be applied to a place learner and a response learner to make the dissociation concrete.

The Neural Basis of Spatial Learning

The neural account of spatial learning is one of the clearest links between a cognitive construct and its cellular substrate. O'Keefe and Dostrovsky reported that individual neurons in the rat hippocampus fire only when the animal is in a particular part of its environment — place cells, each with a place field covering a small region of space — and proposed that the population of such cells constitutes the neural cognitive map (O'Keefe & Dostrovsky, 1971; O'Keefe & Nadel, 1978). The map is completed upstream in the entorhinal cortex, where grid cells fire at the vertices of a regular triangular lattice tiling the environment, supplying a metric — a coordinate system of distance and direction — that place cells and the hippocampus can read out (Hafting et al., 2005).

That this machinery is necessary for allocentric spatial learning, and not merely correlated with it, was shown by lesion. Morris and colleagues found that rats with hippocampal lesions were severely impaired at finding a hidden platform whose location could be recovered only from distal room cues, yet learned normally to swim to a platform that was visibly marked — a double demonstration that the hippocampus is required for place learning specifically and not for the sensorimotor or motivational components of the task (Morris et al., 1982). The result is among the most cited in behavioral neuroscience because it isolates the maplike component so cleanly.

What actually stores the map is activity-dependent synaptic plasticity. Blocking the hippocampal NMDA receptor — the coincidence detector that gates long-term potentiation, the lasting strengthening of co-active synapses — leaves place cells firing but prevents rats from learning the hidden platform, tying the acquisition of a new spatial map to the same plasticity mechanism thought to underlie memory generally (Morris et al., 1986). The map is therefore not innate but built into hippocampal connectivity through experience.

The same architecture operates in humans, where its plasticity is visible at the scale of a career. Structural imaging of licensed London taxi drivers, who must memorize the city's street layout, found enlarged posterior hippocampi relative to controls, with volume scaling with years of navigation experience — direct evidence that intensive allocentric learning remodels the human hippocampus (Maguire et al., 2000). Place-cell, grid-cell, and lesion findings thus align across species into a single account: spatial learning is the acquisition of a hippocampal-entorhinal map whose units code position, and whose integrity determines whether an organism can navigate by place at all (Grieves & Jeffery, 2017).

Measuring Spatial Learning

Two tasks dominate the experimental study of spatial learning, each isolating a different facet. The Morris water maze places an animal in a pool of opaque water with a platform hidden just beneath the surface; escape is possible only by learning the platform's location relative to distal room cues. Learning is read from the falling escape latency across trials and, on a probe trial with the platform removed, from the proportion of time spent searching the correct quadrant. Because the platform is invisible and the water removes local guidance, the task cannot be solved by any non-spatial strategy, which is what made it the standard assay of place learning and the tool that established the hippocampal dependence of that learning (Morris, 1984; Morris et al., 1982).

The radial-arm maze separates two kinds of spatial memory within one apparatus. Arms radiate from a central platform, each baited once; an efficient forager visits each arm exactly once and does not return. Re-entering an already-emptied arm is a working-memory error — a failure to remember where one has been within the current trial — whereas entering an arm that is never baited is a reference-memory error, a failure to learn the stable rule that holds across trials. The dissociation let Olton and Samuelson demonstrate that rats hold an accurate, list-like memory for the places already visited, and it remains the canonical way to separate trial-specific from rule-based spatial memory (Olton & Samuelson, 1976). The demonstration below tracks working- and reference-memory errors as arms are chosen.

Worked Example

Consider how escape latency in the water maze falls as an animal learns the platform's location. Across trials, latency typically decays toward a floor set by swim speed and the distance to the platform, and a standard description is the exponential learning curve L(n) = L_∞ + (L₀ − L_∞)·e^−n/τ, where n is the trial number, L₀ the initial latency, L_∞ the asymptotic floor, and τ the learning time constant in trials.

Suppose an animal starts at L₀ = 60 s, has a floor of L_∞ = 8 s, and learns with τ = 3 trials. The latency then falls 60 s → 45.3 s → 34.7 s → 27.1 s across the first three trials, since each trial multiplies the remaining excess above the floor by e^−1/3 = 0.717. To ask when the animal reaches a performance criterion of L ≤ 15 s, solve 8 + 52·e^−n/3 ≤ 15, i.e. e^−n/3 ≤ 7/52 = 0.1346, giving n ≥ −3·ln(0.1346) = 6.02. The criterion is therefore first met on trial 7: at trial 6 the latency is still 15.04 s, a hair above criterion, and at trial 7 it is 13.04 s.

The time constant sets the whole trajectory. A faster learner with τ = 2 reaches the same 15 s criterion in five trials rather than seven — at τ = 2 the latency is still 15.04 s at trial 4 and first drops below criterion, to 12.27 s, at trial 5 — while a hippocampally impaired animal with a much larger τ may never separate from its starting latency within a session — which is exactly the pattern lesion studies report, and why the shape of the latency curve, not a single trial, is the measure of spatial learning (Morris, 1984; Morris et al., 1982).

Discussion

Spatial learning occupies a privileged place in the science of learning because it is where an abstract cognitive construct — a map in the head — was given a mechanism that can be recorded from directly. Tolman's inference that animals learn spatial relations rather than movement sequences was, for decades, a behavioral argument; the place cell turned it into a claim about identifiable neurons, and the grid cell supplied the metric the map needs (Tolman, 1948; O'Keefe & Dostrovsky, 1971; Hafting et al., 2005). The convergence of behavior, single-unit physiology, lesion, and human imaging on the hippocampal-entorhinal system makes spatial learning one of the best-understood forms of learning at the level of implementation (Morris et al., 1982; Maguire et al., 2000).

The distinction between allocentric and egocentric learning is what keeps the topic from collapsing into a single mechanism. Because the same destination can be reached either by a hippocampal map or by a striatal habit, spatial tasks must be designed to force one solution or the other, and the interpretation of any deficit depends on which system a task actually taxed. This is also why spatial learning connects to the wider study of memory and flexibility: an intact map supports novel routes and rapid detours, a behavioral hallmark of flexible cognition, while a reliance on fixed routes is the spatial signature of its loss. The map, in short, is not only how an animal finds its way but a model system for how the brain represents structured knowledge at all.

Current Directions

The most active current work extends the cognitive map from physical space to knowledge in general. Human functional imaging and intracranial recording have established that the hippocampal-entorhinal system that codes location also organizes non-spatial information — conceptual, social, and temporal relations — along maplike dimensions, suggesting that spatial learning is a special case of a general capacity to represent structured relations (Epstein et al., 2017; Bellmund et al., 2018). Grid-like coding, measured indirectly in humans, appears when people navigate abstract feature spaces as well as rooms, which has reframed the map as a domain-general format for cognition rather than a navigation-specific device.

A parallel debate concerns the geometry of the learned representation itself. Where the classical cognitive map is metric and Euclidean, recent accounts argue that much spatial knowledge is better described as a graph of remembered places and the transitions between them — a topological structure that explains why human distance and direction judgments are systematically distorted, and why people often navigate by a network of familiar routes rather than a survey map (Peer et al., 2021). Reconciling metric and graph-based descriptions, and specifying when the brain uses each, is an open question that current single-unit and modeling work continues to pursue (Grieves & Jeffery, 2017).

Common Misconceptions

Spatial learning is just memorizing a route.

A learned route is only the egocentric form. Allocentric spatial learning yields a maplike representation that supports novel shortcuts and detours the learner never practiced — the very flexibility Tolman used to argue that animals acquire a cognitive map rather than a chain of responses (Tolman, 1948).

Place cells are like a GPS that stores absolute coordinates.

Place cells fire relative to the environment's own landmarks and boundaries, not to fixed global coordinates; a given cell's field can remap entirely in a different environment. The metric of distance and direction comes from entorhinal grid cells, which the place-cell population reads out (O'Keefe & Dostrovsky, 1971; Hafting et al., 2005).

Hippocampal damage abolishes all spatial performance.

It abolishes place learning specifically. Rats with hippocampal lesions cannot find a hidden platform defined by distal cues, yet swim directly to a visibly marked one, because cued approach and egocentric route learning are supported by other systems (Morris et al., 1982).

Glossary

Allocentric frame.

A representation in which locations are coded relative to external landmarks and boundaries, independent of the observer's position; the format of the cognitive map.

Cognitive map.

An internal, maplike representation of the spatial layout of the environment that supports flexible navigation, including novel shortcuts; proposed by Tolman and localized to the hippocampus.

Egocentric frame.

A representation in which locations are coded relative to the observer's own body and movements; the basis of route and response learning.

Escape latency.

The time taken to reach the hidden platform in the water maze; its fall across trials is the primary index of spatial learning.

Grid cell.

An entorhinal neuron that fires at the vertices of a regular triangular lattice covering the environment, supplying the metric of distance and direction for the cognitive map.

Hippocampus.

A medial temporal lobe structure required for allocentric spatial learning; it houses place cells and is enlarged in expert human navigators.

Latent learning.

Learning that occurs without reinforcement and is not expressed until an incentive is introduced; evidence that animals acquire spatial knowledge in the absence of reward.

Long-term potentiation.

A lasting, activity-dependent strengthening of synaptic transmission, gated by the NMDA receptor; the leading candidate mechanism by which the hippocampal spatial map is stored.

Maze learning.

Learning the correct route or location within a maze to obtain reinforcement; the MeSH subtype of spatial learning and the methodological core of animal research.

Morris water maze.

A pool with a hidden submerged platform located from distal cues; the standard assay of place learning and its hippocampal dependence.

Place cell.

A hippocampal neuron that fires selectively when the animal occupies a particular location; the population of place fields forms the neural cognitive map.

Place field.

The circumscribed region of an environment in which a given place cell fires; firing rate is highest at the field's center and falls off with distance.

Radial-arm maze.

An apparatus of arms radiating from a center, each baited once, that separates working-memory errors (re-entering a visited arm) from reference-memory errors.

Reference memory.

Memory for the stable features of a task that hold across trials, such as which arms are ever baited; dissociable from trial-specific working memory.

Response strategy.

Solving a spatial task by repeating a learned body-turn or movement sequence rather than by consulting a map; the egocentric alternative to a place strategy.

Spatial memory.

The retention of learned information about locations and spatial relations; the stored product of spatial learning.

Working memory (spatial).

Memory for locations visited within the current trial or episode, such as which maze arms have already been entered; dissociable from cross-trial reference memory.

Key Researchers

Richard G. M. Morris. University of Edinburgh; he devised the Morris water maze, the dominant assay of rodent spatial learning, and used it to show that hippocampal lesions and NMDA-receptor blockade impair place navigation while sparing cued approach. ORCID - Wikipedia

Edvard I. Moser. Norwegian University of Science and Technology; co-discoverer of entorhinal grid cells, the lattice-firing neurons that supply the metric of the cognitive map, for which he shared the 2014 Nobel Prize in Physiology or Medicine. Wikipedia

May-Britt Moser. Norwegian University of Science and Technology; co-discoverer of grid cells and of the entorhinal microstructure of the spatial map, and a 2014 Nobel laureate for that work on the brain's positioning system. ORCID - Wikipedia

Lynn Nadel. University of Arizona; with John O'Keefe he co-authored The Hippocampus as a Cognitive Map (1978), the monograph that recast the hippocampus as the substrate of allocentric spatial learning. ORCID - Wikipedia

John O'Keefe. University College London; he discovered hippocampal place cells in 1971 and, with Nadel, formulated the cognitive-map theory of hippocampal function, sharing the 2014 Nobel Prize in Physiology or Medicine. ORCID - Wikipedia

Edward C. Tolman (1886-1959). University of California, Berkeley; his latent-learning and place-learning experiments in rats produced the concept of the cognitive map, the founding idea that animals learn spatial relations rather than chains of responses. Wikipedia

Frequently Asked Questions

What is spatial learning?
Spatial learning is the acquisition of knowledge about the locations of places and objects in the environment and how they are arranged, studied in tasks where only location must be learned. Its maplike form allows an organism to find a place again and to take novel shortcuts (Tolman, 1948).

What is the difference between allocentric and egocentric spatial learning?
Allocentric learning codes locations relative to external landmarks in a viewpoint-independent map, while egocentric learning codes them relative to the observer's own body as a route or sequence of turns. Most natural navigation combines the two, and they can be dissociated experimentally (O'Keefe & Nadel, 1978).

Which part of the brain supports spatial learning?
The hippocampus and the surrounding medial temporal lobe are the core substrate of allocentric spatial learning. Hippocampal lesions abolish the ability to find a location defined by distal cues while leaving cued approach intact (Morris et al., 1982).

What are place cells and grid cells?
Place cells are hippocampal neurons that fire when the animal is in a specific location, and grid cells are entorhinal neurons that fire in a regular lattice across space. Together they provide the position code and metric of the cognitive map (O'Keefe & Dostrovsky, 1971; Hafting et al., 2005).

How is spatial learning measured?
The Morris water maze measures it through the falling escape latency to a hidden platform and time spent in the correct quadrant on a probe trial, while the radial-arm maze separates working-memory from reference-memory errors (Morris, 1984; Olton & Samuelson, 1976).

What is a cognitive map?
A cognitive map is an internal, maplike representation of the spatial layout of the environment that supports flexible navigation, including routes never practiced. Tolman inferred it from rat behavior, and it was later localized to the hippocampus (Tolman, 1948; O'Keefe & Nadel, 1978).

Can spatial learning change the human brain?
Yes. Licensed London taxi drivers, who memorize the city's layout over years, have enlarged posterior hippocampi that scale with navigation experience, showing that intensive allocentric learning remodels the human hippocampus (Maguire et al., 2000).

Is the cognitive map used only for physical space?
Recent human evidence indicates that the hippocampal-entorhinal system also organizes non-spatial, conceptual, and social information along maplike dimensions, suggesting spatial learning is a special case of a general capacity to represent structured relations (Bellmund et al., 2018; Epstein et al., 2017).

References