Reinforcement Learning

Learn how to sail safely in dangerous waters to your goal by creating a generation based reinforcement learning model using Q-Learning or SARSA.

What makes the waters dangerous?

There are three types of tiles in the deterministic environment:

1. no waves - every action moves the boat by 0 additional fields
2. yellow waves - every action moves the boat by 1 additional field
3. red waves - every action moves the boat by 2 additional fields

When you use the stochastic environment you can choose the probability of the waves.

Configuration

You can configure:

• learning algorithm
• learning rate (alpha)
• discount factor (gamma)
• number of generations
• exploration and explotation proportion
• value of positive/negative rewards
• environment type (stochastic or deterministic)

Statistics

The program automatically generates, shows and exports some related statistics.

Heatmaps

Q-Values

A .csv file showing the q-value of every action (columns) for each episode (rows).

Rewards/Episode

Steps/Episode

This program was developed as part of a university project.