This paper presents a methodology for autonomous navigation of mobile robots with differential traction in poorly structured environments. The objective of the developed system is to navigate in areas with different types of obstacles could exist to go from one point to another without collision. The navigation methodology uses a probabilistic neuronal network (PNN) as a decision core for control the motion of the mobile robot during its path. The methodology is implemented in the Optimus System, and the results obtained allow validate its performance.