QLearning-RL-Maze

Find 5 treasures and EXIT with minimum steps (QLearning)

Map:

• wall positions: [4,5,7,9,22,23,25,30,31,35,39,43,45,47,49,50,51,53,55,57,58,59,61,65,71,74,80,85,88,90,94,97,100,101,102,104,109,110,111,113,114,119,120,127,128,129,132,134,136,141,142,143,145,151,153,155,157,158,164,166,169,172,176,178,181,183,186,187,190,191,193,195,196,206,211,214,226,229]
• treasure positions: [6, 79, 170, 212, 227]
• exit position : 230

Settings:

• hyperparameters Setting -> EPSILON = 0.9 (randomness of ACTIONS)
  -> ALPHA = 0.1 (learning-rate)
  -> GAMMA = 1 (desire for future rewards: 0 -> ignore future rewards, 1 -> look for high rewards in the long term)
  -> MAX_EPISODES = 1000 (amount of times of walking through the maze)
  
• Reward Setting -> goal_reward (exit found): 5000
  -> wall_punish (hit into the wall): -300
  -> out_punishment (go out of the map): -200
  -> treasure_found: 1000
  -> normal_reward (normal path): -100
  

Best Result with no treasure found

Steps = 60 (at episode 442)
Path:

Best Result with 5 treasure found

Steps = 501 (at episode 80)
Path: