Q-Learning: Solutions for Grid World Problem with Forward and Backward Reward Propagrations

In the area of adaptive and responsive problems, Reinforcement Learning algorithms have made significant progress. This paper presents solutions to the grid world problem using the model-free reinforcement learning method known as the Q-Learning algorithm. The solutions are developed with forward and backward reward propagations under the assumption of a Markov decision process for the grid world problem. This study detail the implementation and comparison of these two reward calculations. The paper also illustrates how an agent interact with the gird environment during both forward and backward propagations and compare their benefits followed by hyperparameter tuning for better understanding of the model’s convergence.