Simulation-Based Simple and Robust Rule Generation for Motion Coordination of Multi-agent System

Abstract

For motion coordination of a multi-agent system, a simulation-based rule generating method is needed. However, previous studies assuming understandabieness of calculated rules for users do not exist. In this paper, a simulation-based simplified and robust rule generation system for multi-agent scheduling problem is proposed using parallel algorithm discovery and orchestration (PADO) and simulated annealing programming (SAP). In addition, a method to extract constraints of rules from a simulation is also proposed. This proposed method is evaluated with an aircraft control problem, and robust rules can be calculated. Moreover, with the method to extract constraints, the average calculation time can be 80 % less than that without the proposed method.

Simulation-Based Rule Generation Considering Readability

“…The experiments were carried out by comparing the proposed method and the one proposed in [ 17 ]. We perform each approach over the course of 10 hours and compare the robustness and readability of the final solutions.…”

“…In this paper, we proposed a combined method of PADO proposed by [ 18 ] and Operator Equalization proposed by [ 19 ] and used in [ 17 ], but incorporating metrics on readability from [ 16 ] on our the rule generation method. By applying our proposed method to the problem of aircraft ground control, it was possible to obtain a set of rules that was both more robust and more readable than the existing rule set, as well as when compared with the rules generated by the former method.…”

“…We base our method on the method proposed in [ 17 ]. An overall picture of the algorithm is shown in Figure 4 .…”

“…When improving robustness and readability is treated as a multiobjective optimization problem, such as in [ 17 ], the problem, as stated in [ 21 ], will fall into local optimal solutions, which undesirably lowers the quality of the solution. In this study, we prevent this by building in optimization of the graph by PADO via a technique called Operator Equalization.…”

“…In addition, in [ 17 ], although a rule generation method for multiagent systems has been proposed, evaluation was not sufficient from the viewpoint of the scale of the problems to evaluate.…”