Exploration in high-risk areas is a topic that has motivated the development of mobile robotics in recent years. Moreover, the incursion of multi-agent systems in this field has opened a lot of solutions and applications. In this paper we propose a strategy of exploration and mapping for multi-robot systems in underground mines where toxic gases concentration (ex.: CO<inf>2</inf>, CO, Sb) is unknown. The principal algorithm is the behavior control which evaluates the status of each agent and makes decisions that maximize system performance and minimize the cost of them. We will use scanning algorithms based on dynamic graph to reduce bandwidth consumption and use of memory. The system has been tested by simulating several situations such as partial loss of communications or agents.