Modelling a Reinforcement Learning Agent For Mountain Car Problem Using Q – Learning With Tabular Discretization

The advancement of Reinforcement Learning (RL) algorithms has shown great success in the field of adaptive and responsive video games. Open AI gym models are open-source visual interfaces for representative environments for modelling robust reinforcement learning tasks. This work aims at solving the mountain car problem involving the "MountainCar-v0" environment used from the OpenAI gym collection framework. Using the Q - learning algorithm, we present a novel algorithm using Tabular-Discretization for solving the considered problem. The gym environment we have considered has a default reward threshold, which is the reward value that an agent must earn before the task gets completed. The proposed model achieved a value close to the threshold parameter set for the experiment. Hyperparameter tuning has been carried out in the latter, to better understand the convergence of the model in terms of the variance of all 4 hyperparameters taken into account- learning rate, decay rate, epsilon, and the discount factor. This proposed method of Tabular Q-Learning is both memory and time efficient in terms of other existing algorithms such as SARSA and Online Q-Learning.