RL Goal Seeker

Goal_seeker.mp4

Goal Seeker — Reinforcement Learning Simulator

Goal Seeker is a lightweight, browser-based reinforcement learning simulator for experimenting with prediction and control algorithms in a grid-world environment. It provides a visual, interactive way to observe how an agent learns to reach a goal using classic RL methods such as Q-Learning.

The project is intentionally simple and framework-free, making it ideal for learning, teaching, and rapid experimentation.

---

Features

• Reinforcement Learning Algorithms

  • Q-Learning (tabular)
  • SARSA (UI-ready; easy to plug in)

• Grid-World Environment

  • Obstacles, empty cells, and terminal goal states
  • Configurable rewards

• Interactive Simulator

  • Step through learning one action at a time
  • Run full episodes at adjustable speed
  • Reset environment and policy instantly

• Live Hyperparameter Control

  • Learning rate (α)
  • Discount factor (γ)
  • Exploration rate (ε)

• Real-time Stats

  • Episode count
  • Steps per episode
  • Last received reward

• Modern UI

  • Responsive layout
  • Light / dark theme toggle
  • Keyboard shortcuts

---

How It Works

1. The Agent exists in a discrete grid environment.
2. At each step:
   • The agent selects an action using an RL algorithm.
   • The environment returns a reward.
   • The algorithm updates its value estimates.
3. Learning continues until the agent reaches the goal state.

The simulator cleanly separates:

• Environment (Grid, Cell)
• Learning algorithm (Algorithm, QLearning)
• Visualization (Visualizer)
• Control logic (Simulator)

---

Running the Project

Option 1: Open directly

Open index.html in a modern browser.

Some browsers restrict ES modules from local files.
If you see errors, use a local server.

Option 2: Local server (recommended)

Clone the repository

• git clone repo-url
• cd project-folder

Start a local server

• npx http-server

Open in your browser

• http://localhost:8000