Run & Catch – Extended Q-Learning (Made by GeFA)

This repository contains a single HTML file (index.html) that demonstrates a Q-Learning-based “Run & Catch” scenario on a 6×6 grid:

• Two Runners (blue) try to survive for 30 seconds.
• Two Chasers (red) attempt to catch both Runners before time runs out.
• A Q-Learning system updates the policy for both teams (Runners and Chasers) each episode.
• LocalStorage is used to save/load the learned Q-tables, allowing the AI to persist across browser sessions.
• A custom HUD shows the game state (time, caught runners, etc.), and a Chart.js line graph tracks the Runners’ Win-Rate over multiple episodes.

---

Main Features

1. Q-Learning for Both Teams

   • Each team has its own Q-table.
   • (\epsilon)-greedy actions; rewards for distance changes, attacking, catching, or avoiding capture.

2. Auto-Restart Episodes

   • Game runs for up to 30 seconds.
   • If both Runners are caught beforehand, the Chasers win immediately.
   • Otherwise, if time expires with at least one Runner free, the Runners win.
   • The code automatically starts a new episode (unless stopped), incrementing an episode counter.

3. Model Persistence via LocalStorage

   • Q-tables are saved after each episode.
   • On reload, the game continues from previously saved Q-tables.

4. HUD & Live Chart

   • Time left, actions taken, episodes, and game status displayed in a panel-based HUD.
   • Chart.js line chart for Runners’ overall win-rate across multiple episodes.

---

Usage

1. Clone or Download this repository (you only need index.html).
2. Open index.html in a modern web browser:
   • Either directly (some browsers might restrict localStorage if loaded via file://).
   • Or, use a basic local server, for example:

     python -m http.server

     then go to http://localhost:8000/index.html in your browser.
3. Click “Start” to begin the first episode:
   • Two Runners and two Chasers will move on the 6×6 grid.
   • The Runners try to survive; the Chasers try to catch them.
4. Observe the Q-Learning in action:
   • At each episode’s end, the result (Runners win or Chasers win) updates the stats.
   • After a brief pause, the next episode starts automatically (unless you click “Stop”).
5. Monitor the HUD:
   • Time left in the current episode.
   • How many Runners are caught.
   • The chosen action (e.g., UP, DOWN, STAY, ATTACK, SPECIAL) for each team.
   • The overall episode count, and in the Chart, the Runners’ win-rate.
6. Stop the process anytime with the “Stop” button. The next run can continue from the stored model.

---

Customizing

• Adjust Q-Learning: Tweak parameters (\alpha, \gamma, \epsilon) in the Brython code.
• Rewards: Edit the distance-based rewards, catching penalty/bonus, or attack/special logic to encourage different AI behaviors.
• Auto-Restart: Switch off AUTO_RESTART or change EPISODE_DELAY_MS in the code to control the downtime between episodes.
• LocalStorage: If you prefer ephemeral training, remove the calls to load_all_q_tables() / save_all_q_tables().

---

Contributing

Feel free to open issues or pull requests to improve the AI logic, add new features, or enhance the visuals. Feedback is welcome!

---