This paper addresses the problem of optimizing the observation of a human scene using several mobile robots. Mobile robots have to cooperate to find a position around the scene maximizing its coverage. The scene coverage is defined as the observation of the human pose skeleton. It is assumed that the robots can communicate but have no map of the environment. Thus the robots have to simultaneously cover and map the scene and the environment. We consider an incremental approach to master state-space complexity. Robots build an hybrid metric-topological map while evaluating the observation of the human pose skeleton. To this end we propose and evaluate different online optimization strategies exploiting local versus global information. We discuss the difference of the performance and cost. Experiments are performed both in simulation and with real robots.