In this paper, we show how neural controllers can be evolved to let robots autonomously develop a form of spatial representation, which allows the discrimination of different places within their environment. In particular, the developed representations consist in patterns of activations of the internal neurons, which are generated integrating sequences of sensory motor states through time. Moreover, the developed representation forms are allocentric. This feature allow robots to successfully recognize a place independently from their initial position. The analysis on the robots' representation system indicates that it can be described as a limit cycle resulting from the transient dynamics between fixed attractor points that alternate while the robot moves in the environment.