The mobile robots are devices with great boom given the possibilities that their utilities offer, and to a greater extent, those freelancers who do not require an operator to perform their functions. In order to consolidate the autonomy it is necessary to generate a system of planning of ways that allows a viable route and as far as possible optimal. This study develops a reactive two-dimensional path planning method with neural networks trained under the reinforcement learning method. The complexity of the scenario between the initial and final point is due to warning and forbidden obstacle zones, and the experimentation is carried out on different neural network architectures, each one as an agent of the learning-by-reinforcement algorithm, being these DQN and DDQN types. The best results are obtained with the DDQN training, reaching the objective in 89% in the validation episodes, although the DQN method shows to be 15.63% faster in its success cases. This work was carried out within the research group DIGITI of the Universidad Distrital Francisco José de Caldas.