Bias and Raising Threshold Algorithm Using Learning Agents for the Best Proportion-Searching Problem

In robotics systems, the best-of-n problem is well-known as a standard problem in which a group of distributed robots is required to make a collective decision on the best among a set of n options without a leader [1]. As a method for the best-of-n problem, BRT algorithm, proposed by Phung et al. [10], can handle many alternatives. In this paper, we consider the best proportion-searching problem, a subclass of the best-of-n problem, where a group of robots need to find its best proportion that offers the optimal state of the group. In our previous research [14], we have introduced roles to individuals, used the combination of individual roles generated in advance as an option for the decision-making based on BRT algorithm. The computer experiments show that the best proportion can be found if the combination of individual roles is appropriately set in advance. However, in some cases where there is no the best proportion among the set of options, the group of robots cannot find out the best one. To this problem, in the present study, we propose a method based on learning agents who can change their role when they judge that the selected option is not the best one.