While handling missions collectively, robots ought to have the option to organize their endeavors, for example, by following through with various sub-jobs, checking various pieces of an objective climate, etc. Throughout the course of recent years, PC researchers have, in this manner, been creating computational models intended to organize the activities and ways of behaving of various robots in a group.
Scientists at Lulea College of Innovation in Sweden have recently presented a new multi-specialist coordination technique that coordinates a closeout-based task with conduct trees, numerical models frequently utilized in software engineering to execute plans. In a paper that was pre-published on arXiv, this method was discovered to effectively organize the actions of multiple robots working toward a common objective, especially when they are completing tasks that span multiple stages.
According to Niklas Dahlquist, one of the study’s authors, “We found a need for a flexible and reactive task allocation architecture for coordinating multi-agent systems.” Because I have previously worked with behavior trees, I believed that combining them with a task allocation scheme could produce an adaptable framework that also makes it simple to incorporate new tasks and agents.
“Future research will concentrate on adding new types of tasks, introducing heterogeneous agents like UAVs, and increasing the complexity of the interaction between task allocation and behavior trees.”Niklas Dahlquist, one of the researchers who carried out the study,
The design created by Dahlquist and his partners depends on a market-based approach, which supports sales. In essence, the team’s individual robotic agents place “bids” through a central auctioning system to estimate the cost of completing various subtasks. After that, this auctioning system looks at these “bids” and gives each agent the best tasks.
Dahlquist explained, “Our architecture also has a local layer, which is a behavior tree, that guides the individual agents to complete an assigned task.” This gives the benefit that there is no requirement for an incorporated unit to have full worldwide information about the framework; everything necessary are the offers from the specialists, and furthermore, it decreases the computational weight.”
Dahlquist et al.’s multi-agent coordination architecture may have significant advantages over other models that have previously been proposed and typically rely only on behavior teams or auction theory. This strategy, for instance, may be simpler to apply to teams consisting of robotic agents with distinct capabilities or missions involving very distinct subtasks if task allocation and execution steps are separated.
According to Dahlquist, “the most notable takeaway from this work is the integration of behavior trees with task allocation to allow for the inclusion of multi-stage tasks.” This gives an adaptable and versatile engineering for handling situations where the particular undertakings are obscure ahead of time.”
Using a number of TurtleBot3 robots, Dalquist and his colleagues evaluated their architecture in a series of lab experiments. Because their method could coordinate the efforts of these robots when they were collectively tackling simple missions, their findings were very promising.
Later on, this new engineering could be applied and tried in more complex situations, including various kinds of robots or a more extensive scope of sub-undertakings. Additionally, it might serve as a catalyst for the development of comparable strategies for coordinating the efforts of various agents.
“Our future works will zero in on consolidating new sorts of errands and presenting heterogeneous specialists, like UAVs, and a more complicated coupling between the undertaking distribution and the conduct trees,” Dahlquist added. “In addition, we are attempting to demonstrate the viability through larger deployments in environments that are more realistic.”
More information: Niklas Dahlquist et al, Reactive Multi-agent Coordination using Auction-based Task Allocation and Behavior Trees, arXiv (2023). DOI: 10.48550/arxiv.2304.01976