BCS_secy_submission_Keshav_Agarwal

Goblet of Fire

A = Death Eater

B= Harry Potter

C= The Cup

🧠 Maze RL Simulation – B Escapes A and Reaches C

In this project, I implemented a Reinforcement Learning (Q-learning) agent B that learns to escape a chaser A and reach the goal C in the given grid-based maze. I have rendered the environment using pygame, and training progress is visualized using matplotlib.

---

✅ Basic Approach

• Environment: The maze is already pre-defined in a text file (V2.txt) where:

  • 'X' represents walls
  • ' ' (space) represents free paths

• My Agents:

  • C: Static target. ( The Cup ) Its position is constant during each episode.
  • B: Learns to reach C using Q-learning and avoids A.
  • A: Always chases B using Breadth-First Search (BFS).

• I have drawn the maze and then initialized random positions for A,B,C

• Movement: All agents can move in the directions up, down, left, right—only if there is no wall.

• ** Defining Q-Learning for B**:

  • States: Grid positions
  • Actions: 4 directions
  • Rewards:
    • +100 for reaching C
    • -100 for getting caught by A
    • +10 for moving closer to C
    • -1 to -10 for poor moves (walls, backtracking, etc.) -also using noise

• My logic for training B:

  • B explores randomly at first (epsilon-greedy), then exploits learned policy to reach C
  • A starts far from B to give it a fair chance
  • Successful runs are counted to compute success rate

-Then I have plotted graphs of success rates using Matplot

⚙️ Assumptions Made

• The maze file V2.txt contains only 'X' for walls and spaces for paths.
• B and C are spawned randomly anywhere in the maze (but C's position stays fixed during the episode).
• A is always spawned kinda farther from B so that B gains some time to learn.
• BFS is used for A to keep its behavior consistent and predictable.

-**Important to type python goblet_of_fire.py train in the terminal to see the plots and the success rate.

---

▶️ How to Run

1. For installing required libraries

pip install pygame matplotlib

✅ 2. For training the Agent ( important )

python your_script.py train

train_agent(1000)

✅ 3. To Run the Simulation (After Training)

python your_script.py