Repository files navigation

Implementation of three gridworlds environmentsfrom bookReinforcement Learning: An Introductioncompatible withOpenAI gym.

$ import gym$ import gym_gridworlds$ env = gym.make('Gridworld-v0')  # substitute environment's name

Gridworld is simple 4 times 4 gridworld from example 4.1 in the [book].There are four action in each state (up, down, right, left)which deterministically cause the corresponding state transitionsbut actions that would take an agent of the grid leave a state unchanged.The reward is -1 for all tranistion until the terminal state is reached.The terminal state is in top left and bottom right coners.

Windy gridworld is from example 6.5 in thebook.Windy gridworld is a standard gridworld as described abovebut there is a crosswind upward through the middle of the grid.Action are standard but in the middle region the resultant states areshifted upward by a wind which strength varies between columns.

Cliff walking is a gridworld example 6.6 from thebook.Again reward is -1 on all transition except those into regionthat is cliff.Stepping into this region incurs a reward of -100and sends the agent instantly back to the start.