Analyzing Collaborative Navigational Maze Behavior in a Multiagent Systems Simulator

We analyze the performance of five navigational approaches for agents searching a maze for an exit. These approaches ranged from non-collaborative (agents with and without recollection of past navigation) to collaborative (agents sharing their map discovery and pursuing joint exploration). In addition to implementing these strategies, we developed a cartographer module to support collaborative map discovery. We ran experiments in a known multiagent systems simulator and analyzed results based on the strategies' effectiveness to maximize the number of agents that reach the exit and their times of arrival. Our study supports the hypothesis that collaborative approaches are more effective when agents share map discovery and explore larger mazes collaboratively.