DinoTracker uses AI to analyse dinosaur footprints

For over a century, dinosaur footprints have been both a valuable discovery and a source of frustration. They offer some of the most direct evidence of animals moving across actual landscapes, yet interpreting them is famously difficult.

A footprint is more than a “stamp” left by a foot. It records mud being squeezed aside, toes slipping, edges giving way and, eventually, erosion altering its shape.

As a result, researchers can examine the same trackway and disagree about whether a predator, a plant-eater or an animal somewhere between the two made it.

A new study suggests that artificial intelligence could help make sense of the confusion.

Researchers have developed DinoTracker, a mobile app that allows users to upload a photograph or even a sketch of a dinosaur footprint and receive an immediate analysis of the type of dinosaur that might have left it.

Why dinosaur footprints puzzle scientists

Footprints do not fossilise in a uniform, predictable manner. Animals with identical foot anatomy can leave very different marks, depending on the sediment, its moisture, their speed and how much their weight distorted the ground.

A track can also change once it has been made. Sediment may become compacted, its edges may break down, and subsequent weathering can remove details or make them appear more pronounced.

For these reasons, conventional footprint research often relies on expert judgement and painstaking comparisons with known tracks.

Many earlier computer-based approaches required researchers to assemble datasets by hand. They also involved assigning tracks to particular dinosaurs, a process that can introduce bias or reinforce existing assumptions.

Training AI to recognise variation

The DinoTracker team was led by researchers at a Helmholtz Research Center in Berlin, working alongside colleagues from the University of Edinburgh.

Rather than forcing footprints into excessively neat categories, the researchers trained their algorithms to recognise the realistic ways in which tracks vary.

The AI learned from nearly 2,000 genuine fossil footprints as well as millions of simulated variations created to imitate natural processes.

These additional versions recreated effects including compression, displaced edges and other distortions. Such changes can make the same type of footprint appear different at different sites.

The system then learned to concentrate on key characteristics that help identify trackmakers, even when a print is imperfect.

The study describes features including the spread of the toes, the position of the heel, the size of the contact area and the apparent distribution of weight as the foot meets the ground.

When AI matches expert judgement

Once trained, the model was tested on its ability to predict which dinosaur was likely to have made a footprint by comparing it with existing fossil tracks.

According to the article, its classifications agreed with those of human experts around 90 per cent of the time, including in cases that are normally contentious.

That does not establish that the AI is “right” in an absolute sense. Footprints can be open to interpretation, and palaeontology often works with the best-supported explanations rather than certainty.

Even so, a system performing at that level could provide a consistent second opinion and draw attention to tracks that merit closer examination.

Dinosaur footprints with bird-like features

One particularly intriguing result involved footprints more than 200 million years old. The AI identified several tracks with unusually bird-like characteristics, resembling prints linked to both extinct and modern birds.

The researchers put forward two possible explanations: birds may have originated tens of millions of years earlier than many timelines suggest, or some early dinosaurs may have had feet that happened to look remarkably like those of birds.

The finding does not resolve the debate, but it adds weight to the possibility that footprints contain signals whose importance has been underestimated.

Scotland’s tracks examined again

The system also revisited puzzling footprints on Scotland’s Isle of Skye. Made roughly 170 million years ago on the muddy shore of a lagoon, the tracks have proved difficult to assign confidently to a particular dinosaur group.

The researchers say the AI suggests that some of the oldest known relatives of duck-billed dinosaurs made the tracks.

If that interpretation is supported, it could change scientific thinking about when and where that lineage began to spread.

Taking AI to real dinosaur tracks

DinoTracker is intended for wider use, not simply as a research demonstration. Footprints are among the dinosaur traces people most commonly encounter in the wild, and an accessible tool could benefit scientists and members of the public alike.

For researchers, it could quickly screen large numbers of tracks and reveal patterns across sites. In education, it makes footprints interactive rather than treating them as purely descriptive material.

During fieldwork, the app also offers a quick means of testing hypotheses on the spot, particularly where interpreting tracks has traditionally depended on the experience of whoever happens to be present.

“This study is an exciting contribution for palaeontology and an objective, data-driven way to classify dinosaur footprints,” said palaeontologist Steve Brusatte from the University of Edinburgh.

“It opens up exciting new possibilities for understanding how these incredible animals lived and moved, and when major groups like birds first evolved.”

Making sense of messy tracks

An app is unlikely ever to settle dinosaur footprint research completely. Tracks are untidy, and the past does not come with labels. Still, the study points to the value of a tool that regards variation as information rather than noise.

If DinoTracker can consistently recognise how real footprints become distorted and still connect them with likely trackmakers, it could accelerate research, broaden participation and put debates on firmer ground.

It may also make the ancient world feel a little more accessible. A footprint captures a moment when an animal came into contact with the ground beneath it.

Reading those moments more clearly could bring us closer to understanding how dinosaurs actually lived, moved and evolved.

Image credit: Tone Blakesley

The full study was published in the journal PNAS.