REINFORCEMENT LEARNING vs. A* IN A ROLE PLAYING GAME BENCHMARK SCENARIO

In a Role-Playing Game (RPG), finding the optimum trajectory of an agent is usually one of the most important objectives. In fact, it becomes a vital point of the game, due to how the path is established (reality or fiction) and the consumed resources (execution time). When classical search algorithms such as A* can be used, they are very useful for computing optimal solutions. Nevertheless, grid-based methods can be computationally expensive, especially for very large environments. Besides, A* based algorithms usually produce aesthetically unpleasant paths and the execution time is higher than evaluating results of the previous learning of the Q-learning algorithm. In this article we evaluate and compare the performance of these classic algorithms, A* and Q-Learning (Reinforcement Learning), on static searching. Simulation results of different simulation scenarios prove that reinforcement learning provides the most optimal path regarding computational cost compared with the A * algorithm depending on the configuration.