This project implements some algorithms, such as Value Iteration and Policy Iteration for calculating the optimal policy of a given Markov Decision Process.

The problem we are trying to solve is a simple 2D grid world problem in which the agent starts off at the grid square (initial state) (3 0) and tries to move to other grid squares located elsewhere with the aim of reaching a terminal state (3 1) or (3 2) while maximizing its gains.

The agent is only allowed actions of moving in up, down, left, right directions by 1 grid square, with probabilities expressed in transitions.csv.

The cost of being in each state is shown in the figure below:

python main.py or via the notebook

transition_file contains tuple (state, action, result-state, probability)

reward_file contains tuple (state, reward)