Methods of Local Behavior Planning for Agents with BDI Architecture

Abstract

This paper deals with the possibility of solving a collective task by a group of intelligent agents with the BDI architecture using three different methods: planning using a global announcement board, coordination through local boards, and the exchange method. The new method proposed in the article is a local boards method based on the modification of the global-board method. As well, the analysis of the exchange method with the proposed formalization and application of the polynomial reduction method was conducted. All three methods were used on a simple simulation of the task of assisted movement on a grid, which explored the applicability of the architecture and methods for a real mobile multi-agent system. The special features of the task are the difficulty of maintaining a common communication channel and the limited computational resources. It was shown that this problem can be solved by introducing local announcement boards, when each agent makes a decision based on access to the advertisements of its neighbors. An analysis of the method based on exchanges has shown that it can be reduced to the formulation of the STRIPS problem. The exchange method allows one to make a search on a smaller graph in comparison to the STRIPS method.