Maze solving problem using q-learning
Chengcong Xu · Applied and Computational Engineering · 2023
In the recent years, a number of research initiatives have employed q-learning. Because of its straightforward logic of assigning a corresponding action to each potential state, it is the most widely used reinforcement learning technique. In our research, we found a way to expedite the training process for the agent. A simple environment was used, a frozen lake. The goal for the agent is to reach the destination by avoiding obstacles after a series of training. A reasonable example was showed in this research. The basic environment and agent were created in Python, and the basic form of Q-learning was utilized. We implemented a q-learning algorithm to solve a 4x4 frozen lake and a complex 8x8 frozen lake. The results showed that the training process takes a long time and is different in more complex environments. We assume there is an exploration-exploitation tradeoff that can speed up the training process. In this way, we define a new parameter, epsilon, which is used to balance the agent during the training process. Also, among methods of exploration-exploitation tradeoff, exponential decay performs better than linear decay.