TetrisRL

• Implement a reinforcement learning model which can play Tetris

Tetris Enviroment

• 10 actions
  • 0: don't move
  • 1: shift left 1 block
  • 2: shift left 2 block
  • 3: shift left 3 block
  • 4: shift right 1 block
  • 5: shift right 2 block
  • 6: shift right 3 block
  • 7: rotate once
  • 8: rotate twice
  • 9: rotate three times
• return
  • pixel(10*20)
  • block_id
  • block_location(x, y)

Version 0

• Policy Gradient Algorithm
• Reward Delay
• Bad Performance

TODO

• change reward function
• reward baseline
• DQN