Abstract
The topic of this paper is the study of two reinforcement learning algorithms, SARSA and Q-Learning. Reinforcement learning is generating significant interest due to its potential applications in various domains such as robotics, gaming, optimization, etc. In addition, reinforcement learning is an interesting object of research from the perspective of theory and practice, as it is related to concepts such as exploration and use, learning and planning, consistency, and stabilization, etc. SARSA and Q-Learning are two of the most well-known and widely used reinforcement learning algorithms, which are based on the evaluation of the value function of states and actions. The aim of this paper is to study the learning characteristics of these algorithms in different scenarios of agent's interaction with the environment. To this end, experiments were conducted in which the agent had to find an optimal path in a 2D matrix containing walls to reach the final position safely. The results showed that SARSA was on average 28.3% faster than Q-Learning.