This paper discusses advantages of dynamical systems approach for the robots to attain intrinsic descriptions in terms of their behaviors. We show qualitative differences between taking views of "external observer" and "internal observer" dealing with the descriptions for the robots. The conventional symbolic approach takes the view of the external observer, in which there exists an observer who looks over the descriptions and try to manipulate them. On the other hand, the dynamical systems approach provides that of the internal observer, in which the descriptions and their manipulations become an unseparable entity. We explain how robots built with the view of the internal observer can be descriptive without having explicit de- scriptions, and how their mental processes of "manipulating descriptions" can be naturally situated in the behavioral context. We describe the problems with reviewing our prior experiments of navigation learning (Tani 1996) which was conducted with using a real mobile robot.