Solving different computational problems with an artificial agent is a challenge, addressed in Artificial Intelligence (AI). Above all, generating autonomy is a growing problem. To achieve this, autonomous agents aim to learn how to accomplish tasks without human assistance. Thus, Complementary Learning Systems (CLS) propose two systems for memory consolidation, based on the functioning of the human brain: the hippocampus (sporadic learning) and the neocortex (general learning) for context inference; this approach is called CIT-SNN. In this article, we propose a strategy for using spiking neural networks (SNNs) to generate an autonomous agent for discrete problems by solving simulated 2D mazes. A practical case is also presented in which an agent was trained to escape from a simulated maze, achieving an average reward of 9.14 in the evaluation stage. In addition, we used the DQN algorithm for comparing this proposal approach.