Scientists from the Artificial Intelligence and Data Science Center at St. Petersburg State University have developed the Softmax algorithm for time synchronization in groups of robots. It allows the system to operate without a central server. Ivan Arkhipov, a specialist at the AI and Data Science Center at St. Petersburg State University, told the "Nauchnaya Rossiya" portal about this.
Modern robotic systems consist of several devices, each performing its own task, but together they solve a common one. For example, during firefighting, one robot detects the fire, a second directs the extinguishing mixture, and a third protects adjacent structures. In warehouses, several robots move and arrange boxes.
In such systems, the order of actions is critically important. Usually, it is controlled by a central server. But if it fails, decentralized algorithms are needed. These are based on each device operating according to its internal clock and exchanging readings with others. The problem is that these clocks can go out of sync due to temperature or other factors. Then the order of actions is disrupted.
Previously, attempts were made to solve the problem using average or maximum time. But the average value often leads to non-monotonicity – time can "roll back," which breaks logs and cause-and-effect relationships. The maximum guarantees monotonicity but leads to infinite acceleration: one node catches up with another, which speeds up in response, and so on in a loop.
The Softmax algorithm is an intermediate function between the average and the maximum. According to Arkhipov, it's a compromise: on the one hand, there are far fewer situations where local time has to be reduced, and on the other hand, monotonicity is maintained due to the maximum.