From Russia the “artificial brain” to guide robots in the crowd

In the not too distant future it will not be difficult to come across groups of robots walking quietly on the streets of our cities, taking the subway or other public transport. However, the need to "instruct" them to move among the crowd without harming human beings is beginning to become increasingly important. A group of researchers from the Department of Cyber ​​Engineering at the National University of Science and Technology MISiS in Russia conducted experiments to try to replicate the same behavior that people have in crowds on a fleet of robots. The result was to have created an artificial neural network capable of being able to imitate the human brain in similar situations .

Our brain is able to spontaneously choose the best path

Have you ever observed the chaotic movement of people in the crowd? Surely it will have happened to everyone, at least once in their life, to impact someone else's path and not know which way to go when this happens. In reality this happens "because of" our brain. When we focus our attention on someone in particular we tend to focus too much on that person's path. If, on the other hand, we walk selflessly, without giving too much loss to other people, this is much less likely to happen. Our brain, in fact, uses the so-called "intuitive knowledge" to unconsciously predict all the movements of people in the crowd, allowing us to choose the optimal path .

Small robots as a case study

The experiments carried out by the scientists of MISiS University involved as many as 55 tiny circular robots (35 mm in diameter) of plastic, printed with the 3D printing technique. Thanks to an elementary electric circuit inside them, the robots are able to vibrate in a range of frequencies between 50 and 100 Hz. By providing an appropriate power supply, they are able to describe random and chaotic movements inside. of a plastic container, thanks to the vibrations of the engine.

Photograph of the robot used to build the artificial neural network. Source: National University of Science and Technology MISiS.
The image shows a photograph of the robot (a) and the control circuit that allows the robot to move by vibrating (b). Source: National University of Science and Technology MISiS.

Cameras to record the movements of robots and train the neural network

During the experiments, the researchers used three different plastic containers with varying diameters (20 to 40 cm). In the three operating conditions, by also varying the coefficient of friction between the robots and the bottom of the container, a camera recorded the trajectories described by the robots. These were used to train DeepLabCut , an artificial neural network based on algorithms made available by the OpenCv Library , a multi-platform software library used for real-time computer vision applications. What the scientists have created is a real "artificial brain" able to define the trajectories and make them available to a fleet of robots, to guide them through the crowd .

The three different plastic containers used for the experiments. From left to right, from the smallest to the largest. Source: National University of Science and Technology MISiS.
The three different plastic containers used for the experiments. From left to right, from the smallest to the largest. Source: National University of Science and Technology MISiS.

The developed neural network will allow scientific groups to significantly simplify the study of physical processes in dense clusters of chaotic moving particles and can be used as a final product. Measuring all the coordinates and speeds of the robots will allow for an exhaustive description of the processes occurring in the crowd, including information on phase transitions and grouping of robots .

Nikita Olekhno, PhD student

Nikita Olekhno, one of the researchers on the team, explained. Furthermore, during the study, the mean square displacement of each robot was also analyzed, starting from the initial position to the final one. Using different lubricants to vary the friction, it was found that for lower densities the robots move less smoothly, while for higher values ​​more natural movements are obtained.

The trajectories of some robots recorded during movement. Source: National University of Science and Technology MISiS.
The trajectories of some robots recorded during movement. Source: National University of Science and Technology MISiS.

The artificial brain can calculate the trajectories of robots in the crowd

The results obtained appear to be promising in order to allow a computer, perhaps remotely, to be able to manage the movements of robots in the crowd. The most surprising thing is that an artificial intelligence may be able to define "randomly precise" trajectories, just like our brains do when we walk among people. The artificial neural network, in fact, can be used to create algorithms for the remote control of fleets of drones . Furthermore, the scientists are optimistic about further improvements thanks to which they can make real-time simulations of the fleet that is being controlled. This would lead to the creation of a Digital Twin (ie a virtual replica) of the entire system. By doing so, it will even be possible to replicate the movements of robots in real time, paving the way for a new frontier of interaction between humans and machines.

The article From Russia the "artificial brain" to guide robots in the crowd comes from Tech CuE | Close-up Engineering .