Abstract

The problem with the Multi-Agent Systems (MAS) is the uncertainty associated with the environment being operated under. The agents can sense the part of environment which lies in its immediate neighborhood. However, these agents are ignorant about the complete information about their world. Hence, it becomes very tedious to avoid any unforeseen hurdles which an agent might face during its future course of actions. This study propose a novel idea to approach the multi-agent planning problems by distributed learning" of the environment. In distributed-learning each agent learns an individual model and shares the results with each other. A solution to this problem has been proposed by first learning the environment i.e. Possible hurdles, free paths, etc. and then finding the plan which when followed will lead each agent to its goal state.