Small single-file browser demo that visualizes a tiny reinforcement learning agent training in a grid world.

Run locally

1. Start a simple HTTP server from the repository root:

python3 -m http.server 8000 --directory .

2. Open the demo in your browser:

http://localhost:8000/microrl.html

Demo

Controls

• Pause / Resume — pause the simulation
• Reset — restart environment and agent
• Speed slider — how many train steps per frame
• Q-map checkbox — overlay per-cell max Q-value heatmap

What I added

• Q-value heatmap overlay (toggleable)
• Smooth agent motion and simple collect/death visual cues
• A small Python script test_rl.py that runs a tabular Q-learning equivalent to validate learning outside the browser

Python test

Run the included Python test to see equivalent training in a headless mode:

python3 test_rl.py --episodes 500

The script prints per-episode rewards and writes rewards.csv with one reward per line.

Files

• microrl.html — main browser demo
• test_rl.py — minimal Python Q-learning test
• LICENSE — MIT license
• CONTRIBUTING.md — short contributing notes

Notes

• microrl.html is ~10KB and under the 50KB threshold; no further shrinking applied.