This Q-Learning code for MATLAB has been written by Mohammad Maghsoudi Mehrabani. It trains an agent to find the way from start point to goal point through a 20x20 maze.
- Actions include turning and moving through the maze
- The agent earns rewards from the environment under certain conditions
- The agent’s goal is to maximize the reward
Base map is stored in maze.txt included Start Point, Goal Point, Walls and Obstacles.
- Goal = 200
- Hole = -10
- Oil = -5
- Free Cells = -1
Just open Main.m file and click on Run button
And Following convergence of the algorithm, MATLAB will print out the best path to the goal
-
Optimal Path which shows the best path from start point to goal point - the best path is shown in below:
-
Figure 1 : Average Cumulative Rewards of each Episode of 30 Runs vs. Episodes
-
Figure 2 : Average Steps of each Episode of 30 Runs vs. Episodes
