The observation and recognition of complex scenes can benefit from the use of multiple mobile cameras. In this paper we study a fleet of mobile robots where each robot controls the point of view of its embedded camera. The objective is to manage the cooperation between the robots to find a joint position that maximizes the joint observation of a scene, defined as the activity of one person. It is assumed that the robots can communicate but have no map of the environment and no external localisation. This paper presents a spatial concentric modeling of the environment well adapted to the navigation of two-wheeled non-holonomic robots. To reduce the complexity of finding the best solution in the space search, we propose an incremental mapping based on this model and some heuristics to search for the optimal observation in an online context. Experimental results in simulation are presented that show in particular the anytime aspects of the proposed algorithms.