Simple example using Q-learning, a foundational reinforcement learning (RL) algorithm.

Vibe-coded with OpenAI's ChatGPT.

• Example: Teaching an Agent to Reach a Goal in a Grid
• Installation
• Run the animation

Example: Teaching an Agent to Reach a Goal in a Grid

The environment is a 4x4 grid where:

• The agent starts in a random position.
• There’s a fixed goal cell that gives a reward.
• Every step has a small penalty to encourage fast learning.
• The agent can move: UP, DOWN, LEFT, RIGHT.

S . . .
. . . .
. . . .
. . . G

• S = Start (random)
• G = Goal (reward = +10)
• Each move = -1 reward
• Goal ends the episode

Note: diagonal moves are not allowed.

• Simplicity: Fewer actions mean a smaller Q-table and simpler policy learning.
• Clarity: Easier to visualize and debug step-by-step movement.
• Tradition: Many tutorials and textbooks use 4-action agents for learning.

Installation

Tested on Ubuntu Desktop 22.04.

First clone this repo.

sudo apt install git git-lfs

git clone git@github.com:guynich/rl_q_learning.git

Create a virtual environment and install packages.

sudo apt install -y python3.10-venv
sudo apt-get install python3-tk

cd
python3 -m venv venv_rl
source ./venv_rl/bin/activate

cd rl_q_learning

pip install --upgrade pip
pip install -r requirements.txt

Run the animation

cd
source ./venv_rl/bin/activate

cd rl_q_learning
python3 main.py

Example run with 100 episodes. The blue square is the goal.

Reload this web page to restart the animation.

Next steps

Great demo for Deep Q Learning : https://projects.rajivshah.com/rldemo/