p-robot / gridworld

Reinforcement learning example, solving the gridworld problem

Gridworld example using different SARSA agents

Reproduction of windy gridworld example by Sutton and Barto (1998) using a range of SARSA agents that learn in different ways.

The figure below is a reproduction of the windy gridworld example of Sutton and Barto (example 6.5, p130 in this book). The figure (figure 6.11 in the linked book) can be reproduced in Python3 using the enclosed code.

Prerequisites

This script was written using Python3.6 and the Python modules described in requirements.txt. It is recommended to use a virtual environment when running this example. A virtual environment can be created and activated using venv in the following manner:

python3 -m venv venv
source venv/bin/activate

pip install -r requirements.txt

Usage

The windy gridworld example can then be run (after activating the virtual environment) using:

python gridworld_tabular.py