Computer scientists are investing a lot of effort in developing computing models that allow coordinating the actions of different robots that have to work as a team. Recently, a group of researchers from the Lulea University of Technology, in Sweden, has found one of the most perfect ways to coordinate various agents with a common goal. And without the intervention of people.
Their experiment, published as an article in the arXiv magazine, shows the need “for a task assignment architecture that is flexible and reactive,” as indicated by one of the authors, Niklas Dahlquist. This scientist remembers that he had previously dedicated himself “to behavior trees,” and that, for this reason, he thought about “combining them with a task attribution scheme” to achieve his purpose.
The system that Dahlquist has designed together with his university colleagues shows an approach based “on the market”, based on the operation of “auctions”, however shocking it may be. The robots individually calculate the cost of certain actions that should be completed and launch “offers” through a general communication tool.
The mission of this instrument is to take into account the “bids” and distribute tasks among the set of machines. The distribution is carried out considering which robot is optimal for each order, since they may have different characteristics or be in an inappropriate situation. The Lulea teachers' solution has the advantage of operating “without a centralized unit.”
Thus, you can have a “global knowledge” of what is happening through the information issued and received by the agents that make up the team. Those responsible for this innovation are convinced that their method can be applied on a large scale. Likewise, they maintain that it would be useful for tackling activities “that were unknown beforehand,” emphasizes Niklas Dahlquist.
So, no matter how heterogeneous the group of robots recruited for a mission may be, the coordination mechanism explained in the arXiv article, based on tests carried out with TurtleBot3 devices, would serve to free flesh-and-blood technicians from the obligation of constantly thinking about which machine is entrusted with a task.
