Skip to content
NatureHQ

Senses and abilitiesmechanism

Path integration

Pick up an ant that has just found food, put it down a hundred metres away, and it will run the exact course home was from where you took it — into empty ground.

Keeping a running total of where home is, updated continuously from your own movement. An animal doing it can be picked up mid-journey, put down somewhere else, and will still run the direction and distance home was from where it was standing.

A desert ant leaves its nest and searches a scorching salt pan on a wandering, looping route that may cover a hundred metres and cross its own path repeatedly. When it finds a dead insect it does not retrace any of that. It turns and runs almost straight for home. To do that it must have been keeping track the whole time — adding up direction and distance as it went, and holding the result as a single vector pointing back to the nest. That is path integration, and the experiment that shows it is a computation rather than a memory is the displacement: pick the ant up at the moment it finds food, carry it to unfamiliar ground, and it runs the vector anyway. It heads in the direction home would have been from where it was taken, for the distance home would have been, and then — arriving nowhere — begins a systematic search. The two halves of the vector come from different senses. Direction comes largely from the polarisation pattern of the sky. Distance comes from the legs: ants given stilts overshoot and ants with shortened legs stop short, which is as direct a demonstration of a mechanism as behavioural biology offers. And because it is a running sum, error accumulates. Path integration alone never quite gets an animal home, which is why it is always the first stage of a return and never the last.

Developed coverage · 52% complete · reviewed 2026-09-03

What this page covers

Demonstrated in desert ants and other insects, in spiders, in rodents and in humans. The most precise measurements come from ants foraging on ground with almost no landmarks.

Often confused with: Retracing a route, which is what path integration is not: the animal computes a straight line home it has never walked; Dead reckoning at sea, which is the same idea but done deliberately with instruments

Quick facts

What it computes
A single vector home, updated continuously from the animal’s own movement
Direction from
The polarisation pattern of the sky, in insects
Distance from
Stride count in ants; optic flow in bees
Its limit
Error accumulates, so it gets an animal close and never all the way

A sum, not a memory

Diagram

The route out, and the line home

Schematic. The outbound path is illustrative; the home vector is the point.

A winding search, and a straight line it never walkedSchematic. The outbound path is illustrative; the home vector is the finding.nestfoodhome vectorNow move the animalreleased heresame direction,same distancethen searches — home is not thereAn animal retracing its route would have repeated the wandering path instead.
The same explanation in words

A nest marked at one corner and a wandering outbound path leaving it — looping, doubling back, and crossing itself several times before reaching a food item. From the food, a single straight arrow runs directly back to the nest, labelled as the home vector. A second panel shows the displacement test: the same animal is picked up at the food and released at a distant point, and from there it runs an arrow of the same direction and the same length as the home vector — which now ends in empty ground, where it begins a looping search pattern. A note explains that an animal retracing its route would have produced the winding path again, and that only a running computation produces a straight line the animal has never walked.

A desert ant walking a winding search does not retrace it. It keeps a running total of where home is and, the moment it finds food, runs straight there — even if you move it first.

Established

Specialists would state this without hedging. Multiple independent lines of evidence agree.

Foraging desert ants continuously update a home vector from self-motion information, and when displaced after an outbound run they travel the direction and distance of that vector rather than retracing the outbound path, with a characteristic systematic error indicating an approximate rather than exact computation.

Who this applies to
Desert ants foraging on featureless salt pans, where the mechanism operates with almost no landmark support.
Studied in
Cataglyphis
Why we rate it this way, and what the caveats are
EstablishedHigh confidence

Displacement makes the prediction unambiguous — a retracing animal and an integrating animal do visibly different things — and the error signature matches a specific computation rather than being noise.

How far it can be extended

Path integration is demonstrated in other insects, in spiders and in mammals; the desert ant is where it is measured most precisely.

Caveats

  • Error accumulates with distance, so path integration alone cannot bring an animal home precisely; the final approach uses landmarks or a systematic search.
  • Studied where landmarks are almost absent, which is what makes it visible and also what makes its everyday weighting elsewhere unclear.

Still unanswered

  • How the running vector is stored and updated neurally, which is understood in outline and not in detail.

Last reviewed 2026-09-03

The evidence (2 studies)

How we know

Letting an ant wander, then moving it, and watching where it runs

Does an ant find its way home by retracing its steps, or by keeping a running total of where home is?

Desert ants foraging on a featureless salt pan were allowed to walk outbound routes of controlled shape — sometimes forced into two straight legs meeting at a known angle. On finding food, each ant was picked up and released on unfamiliar ground some distance away, where none of its own outbound track existed. The direction and length of its subsequent run were recorded.

What happened

Each ant ran in the compass direction of home as computed from its outbound route, for approximately the right distance, then stopped and began a systematic search. The headings carried a small consistent error whose shape matched a specific approximate computation rather than exact trigonometry.

What it shows

That the ant is integrating its own movement into a home vector — direction and distance — continuously as it goes. The error is the most informative part: a remembered route would not go wrong in that particular way, and an exact calculation would not go wrong at all.

What it does not show

It does not show how the vector is stored or updated in the nervous system. It also describes an animal in a habitat almost without landmarks; where landmarks exist, they dominate, so this is the mechanism laid bare rather than the mechanism as usually used.

The controls — what makes this evidence rather than a story
  • Displacement onto ground the ant had never crossed, so retracing is physically impossible.
  • Outbound routes of controlled geometry, so the predicted home vector can be computed in advance.
  • A test area with no landmarks worth speaking of, which is why this species was chosen.

From Path integration in desert ants, Cataglyphis fortis

The systematic error in those homeward runs is the most informative part of the result, and it is easy to skip past. An animal replaying a remembered route would not go wrong in a consistent direction. An animal doing exact trigonometry would not go wrong at all. A consistent, predictable error is the signature of an approximation — a rule of thumb that is cheap to compute and good enough over the distances an ant actually travels.

Words used here
Home vector
The single direction-and-distance pair an animal maintains, pointing from wherever it is back to its starting point.
Systematic search
The expanding loops an animal walks when its home vector runs out and home is not there. It is a search pattern, not confusion.

How an animal measures how far it has come

One of the most direct manipulations in the whole of behavioural biology: change the legs.

How we know

Ants on stilts

How does a desert ant know how far it has walked, in a landscape with almost no landmarks?

Saharan desert ants were trained to walk a set distance from their nest to a feeder. At the feeder — after the outward journey was already complete — some ants had lengths of pig bristle glued to their legs as stilts, and others had the ends of their legs trimmed. Then they were allowed to head home. If an ant measures distance by counting strides, a longer stride should carry it too far and a shorter one should leave it short.

What happened

Stilted ants marched straight past the nest and began searching too far away. Ants with shortened legs stopped short and searched too early. The group that walked both ways on modified legs got it right.

What it shows

Desert ants estimate distance from the number of strides they take. The group that made both journeys on modified legs is what clinches it: they recalibrated, so the mechanism is counting strides, not sensing leg length.

What it does not show

It does not mean ants count in any deliberate sense, and it does not describe how other ants navigate. This is one desert species that lives where there is nothing to look at; ants in cluttered habitats lean far more on landmarks and pheromone trails.

The controls — what makes this evidence rather than a story
  • Unmodified ants walked the same route to establish the correct homing distance.
  • A second group was modified *before* the outward journey, so they walked out and back on the same legs.
  • The terrain was flat and featureless, removing landmark cues.

From The ant odometer: stepping on stilts and stumps

Diagram

Longer legs, longer strides, wrong answer

Schematic. Distances are indicative of the effect, not measured values.

Same number of strides, different stride lengthSchematic. Distances show the effect, not measured values.Normal legsfoodstops at the nestOn stiltsfoodovershoots, then searchesLegs shortenedfoodstops short, then searchesvertical mark = the nestOne full trip on the altered legs restores accuracy.
The same explanation in words

Three ants shown returning along the same homeward direction from the same food site. The first, with normal legs, stops at the nest. The second, with stilts glued to its legs, takes longer strides and stops well beyond the nest before beginning to search. The third, with shortened legs, takes shorter strides and stops short of the nest. A note explains that all three ran the same number of strides, that only the length of each stride differed, and that after one complete outbound trip on the altered legs each ant returns to normal accuracy — because the outbound count and the homeward count now match again.

Give a desert ant longer legs and it overshoots home; shorten them and it stops short. Distance is measured from strides — which is not the same as an ant deliberately counting anything.

Established

Specialists would state this without hedging. Multiple independent lines of evidence agree.

Desert ants whose legs were lengthened with stilts or shortened after an outbound run travelled correspondingly too far or not far enough on the homeward vector, and returned to normal accuracy once they had made a full outbound trip on the altered legs, indicating a stride-integrating odometer rather than an optic-flow or energy-based measure.

Who this applies to
One desert ant species, walking on open ground.
Studied in
Cataglyphis fortis
Why we rate it this way, and what the caveats are
EstablishedHigh confidence

The manipulation is unusually direct and its prediction quantitative: the overshoot and undershoot scale with the change in leg length.

How far it can be extended

Bees measure distance by optic flow instead, which is a direct demonstration that this mechanism is not general even among insects.

Caveats

  • "Counting" is the wrong word: nothing shows an ant representing a number, only that stride information is integrated into a distance estimate.
  • Honey bees solve the same problem with optic flow, so this is one solution among several.
  • Altering legs alters gait and load as well as stride length, though the recalibration after one full trip argues against those being the cause.

Still unanswered

  • How stride information is combined with optic flow in species that appear to use both.

Last reviewed 2026-09-03

The evidence (2 studies)

Not every animal solves it this way, and that is the useful part. A honey bee measures distance by how fast the world flows past its eyes: fly a bee down a narrow tunnel with patterned walls and it reports having gone much further than it has, because the optic flow was greater. Two insects, two entirely different odometers, and neither is the obvious answer of measuring time or effort.

The two ways of measuring a journey

Why it is never the whole answer

Because path integration is a sum of estimates, every step adds a little error and nothing removes it. Over ten metres that hardly matters; over two hundred it puts an ant a body-length or two out, which is enough to miss a nest entrance a few millimetres across. So the vector is a way of getting close, and something else finishes the job: the systematic search, and — where there is anything to see — the panorama.

Landmark memory works differently from a vector, and in some ways more simply. An insect can store a coarse picture of the skyline as seen from the nest entrance and then move so as to reduce the mismatch between what it sees now and what it stored. That is not a map and it involves no computation of position: it is a matching process, and it works beautifully within sight of a familiar place and not at all outside it. The two systems together are what get an animal home — a vector for the distance, a remembered view for the last few metres.

Where this fits

What is still unsettled

  • How is the home vector stored and updated in the nervous system?

    Why it matters: The behaviour is characterised precisely enough to model, which makes it one of the best places to ask how a small brain holds a continuously changing quantity.

    What would settle it: Recording from the central complex of insects during actual outbound and homeward runs, which is now beginning to be technically possible.

  • How do animals weight path integration against landmarks when the two disagree?

    Why it matters: Both systems run at once in most species, and conflict experiments give inconsistent answers across studies and conditions.

    What would settle it: Systematic conflict experiments varying the reliability of each cue, rather than pitting them against each other at full strength.

  • How general is it in vertebrates?

    Why it matters: Path integration is described in rodents and in humans, but the measurements are far coarser than in insects and the mechanism much less clear.

    What would settle it: Displacement designs in vertebrates as clean as the ant experiments, which are hard to arrange without also displacing every other cue.

Claims about this, checked

Things people have heard, and what the evidence actually supports.

The research behind this page

2 studies, newest first. Each one has a page explaining what it found and what it could not show.

2006Science

The ant odometer: stepping on stilts and stumps

Ants on stilts overshot the nest; ants with shortened legs stopped short.

1988Proceedings of the National Academy of Sciences

Path integration in desert ants, Cataglyphis fortis

Ants ran home along a vector close to the true home direction and distance, with a small systematic error whose form matched a specific approximate computation rather than exact trigonometry.

This page is a stop on a longer route

A guided journey reads several subjects in a deliberate order, with an argument for why one follows another. You can join in the middle.

Where to go from here

Each of these follows from something on this page — a relationship in the evidence, a claim people ask about, or the next mechanism along.

How complete this page is, and what it is still missing

NatureHQ publishes its own gaps. This page is at 52% completeness against what we would call a finished subject, and was last reviewed on 2026-09-03. It carries 2 claims and answers 15 mapped search questions.

  • no research from the last few years is attached — check for newer work
  • How the nervous system carries it out — the insect central complex — is named as an open question rather than described, and it is now a substantial literature.
  • Landmark panorama matching is explained in prose and deserves its own treatment with a figure.
  • Vertebrate path integration, including the human evidence, is mentioned only in passing.