lewjae / micromouse_rl Goto Github PK

Navigating in unknown real world is a key challenge in autonomous vehicle or mobile robot application. In this project, the problem is simplified as a robot navigating in an unknown maze and finding its optimal path.

The scope of the project is to develop a motion planning algorithm that enables a robot to explore an unknown maze (environment), to learn the maze layout, and then to find its fastest path from a corner of the maze to its center.

Used Reinforcement Learning to learn an unknown maze and to find its the fastest path.

To run the program: python tester.py test_maze_01.txt

To display the maze: python showmaze.py test_maze_01.txt

• "Multiple steps at one move" has not been implemented.
• The number of training steps is set by training_end = 40000 in robot.py. It must be less than 99000. Larger it is (=more traing), the algorithm produces more consistent an optimal result.
• Parameters such as alpha, gamma and epilson have not been optimized.