Digging under the neural network: an archaeology robot will scan ancient cities to a depth of 10 meters

Russian engineers have created and successfully tested a robot for archaeological excavations. He independently examines the work area using ground-penetrating radar and other sensors, and based on the data obtained, artificial intelligence forms a three-dimensional map of the object of interest to historians at a depth of up to 10 m. The development will save specialists from the time-consuming preparatory stage. According to experts, the technology is convenient to use, but so far it can be used mainly to study individual buildings. For example, when exploring an ancient Egyptian city, where the ruins of some buildings may overlap with others, it will be much more difficult to train a neural network to navigate such a complex structure.

Robot Archaeologist

MIPT scientists have successfully tested the Russian robotic complex TerraCognitaBot. He can independently explore the territory of future excavations using ground-penetrating radar, metal detector, GPS and computer vision. Then the onboard neural network links the collected data to the exact coordinates and builds a three-dimensional underground map of the site with its features. Now this preparatory work is being done by a human.

— The robot will save a person from routine tasks and physical work in difficult conditions. It follows a preset route, simultaneously collects data from several devices and automatically links them to coordinates. The operator can control the operation of the complex and focus more on analyzing the results," said Oleg Bulichev, senior researcher at the Laboratory of Cognitive Dynamic Systems at MIPT.

In the future, the developers plan to train AI to determine the nature of underground objects, such as the remains of a wooden log cabin or a human skeleton. The technology can also be used to search for and identify specific items of historical value. The robot has already proven itself in difficult conditions: in the rain, in tall grass and with strong radio interference.

Manually searching for objects underground is a difficult and painstaking job. An operator with a ground-penetrating radar passes every meter of the site, keeps a steady pace and simultaneously records the coordinates of the location. It is almost impossible to do this manually without mistakes in bad weather, on difficult terrain and in conditions of radio interference.

The power of the ground—penetrating radar antenna is 400 MHz. This is enough to scan to a depth of 10 m. Computer vision helps the robot to keep a set direction of movement, GPS and motion sensors record its position, and timestamps allow you to synchronize data collected by different devices.

​​Sherds and bones: AI helped restore the composition of medieval Russian ceramics

Combining the methods of physics and machine learning in archaeology will reveal the secrets of history.

Tests of the robot archaeologist

Field tests of the complex took place in difficult conditions: It was raining, the grass exceeded the height of the robot, and the equipment was exposed to electronic interference. At the same time, the data was collected in full and with the necessary accuracy. In a straight line, the robot was guided by odometry data, a method for estimating changes in the position of a moving object using data from internal sensors. The course was also adjusted by the neural network. Noticing a specially installed landmark cone, the car braked smoothly.

During the movement, the complex simultaneously collected three streams of data: reflections from the ground from ground-penetrating radar, reaction to metal and GPS coordinates. The timestamps made it possible to synchronize them with each other. The GPR profiles were combined into a single detailed 3D model of the site.

— We were working at an archaeological site where there are traces of burnt wood. Such objects provide a strong contrast to GPR data. We assume that there are also remains of wooden structures, stoves and fireplaces underground," Oleg Bulichev added.

The team is currently processing the collected data and training the neural network to automatically distinguish debris — rocks, roots, and voids — from real targets.

— The development looks promising. Today, an operator with ground-penetrating radar goes through the site manually, monitors the pace and simultaneously records the coordinates. In difficult conditions or with radio interference, it is almost impossible to do this without errors, and technology solves this problem because the robot is guided by computer vision," said Evgeny Vishnevsky, NTI expert on new materials and technologies.

Further testing in large and diverse territories will show how ready the system is for serial use, he added.

As Galina Belova, an Egyptologist and scientific director of the Center for Egyptological Research of the Russian Academy of Sciences, explained to Izvestia, such robotic methods cannot be used everywhere. For example, there is a lot of garbage at the excavations of ancient cities or cemeteries in Egypt.

— In Memphis, where we are conducting excavations, there is almost a two-meter layer of metal debris. And no metal detector or robot will be able to work in such conditions. We tried to use this technology, and it didn't work out. It all depends on the place where it is used. Where there's just a layer of earth, it can be beneficial," she said.

According to her, it is very difficult, and sometimes impossible, to teach a neural network to distinguish between necessary objects and garbage. In the conditions of an ancient city, where one building is superimposed on another, the use of such technology loses its meaning.

According to Natalia Panasyuk, a senior lecturer at the Department of General History of the Faculty of Humanities and Social Sciences of the RUDN University, a robot can be used to collect data, but a human should be involved in interpreting and mapping them.