Gridworld

Simple implementation of text-based Gridworld game. Intended for use with reinforcement learning algorithms.

As described in this blog post: < http://outlace.com/Reinforcement-Learning-Part-3/ >

Basic usage:

import Gridworld as gw

Initialize game state using initGrid (initializes stationary grid, all items are placed deterministically), initGridPlayer (initializes player in random location, but keep wall, goal and pit stationary), or initGridRand in which all pieces are initialized to random locations on the board.

state = gw.initGridRand()
gw.dispGrid(state)

array([['+', '-', 'P', ' '],
       [' ', 'W', ' ', ' '],
       [' ', ' ', ' ', ' '],
       [' ', ' ', ' ', ' ']],
      dtype='<U2')

"P" represents the player, "W" is the wall (an obstacle), and "-" is the pit (a trap that gives a negative reward) and "+" is the goal which gives positive reward and wins the game.

The player makes moves with the makeMove(state, move) where state is the game state and move is an integer 0,1,2, or 3 which translates to 0 = up, 1 = down, 2 = left, 3 = right.

state = gw.makeMove(state, 1) #Down 1 unit
state = gw.makeMove(state, 1) #Down 1 unit
state = gw.makeMove(state, 2) #Left 1 unit
state = gw.makeMove(state, 2) #Left 1 unit
state = gw.makeMove(state, 0) #Up 1 unit
state = gw.makeMove(state, 0) #Up 1 unit

print('Reward: %s' % (gw.getReward(state),))
gw.dispGrid(state)

Reward: 10





array([[' ', '-', ' ', ' '],
       [' ', 'W', ' ', ' '],
       [' ', ' ', ' ', ' '],
       [' ', ' ', ' ', ' ']],
      dtype='<U2')

P player disappears when it encounters the goal or pit.