carrom-ai

Reinforcement learning agent to play the South-Asian tabletop board game Carrom (similar to billiards). CCS 1L project authored by Rohil Shah. See video and summary below for more details.

Carrom.Final.Presentation.Edited.mp4

Auto-generated Video Summary:

1. Understanding Carrom:

   • Brief overview: A two-player tabletop game involving flicking a striker to pocket coins.

2. Project Structure:

   • Simulation:

     • Leveraged an open-source simulator from IIT Bombay.
     • Written in Python, using Pymunk for 2D physics and Pygame for graphics.
     • Customized the open-source code to suit my project's needs.

   • Hand-Coded Algorithms:

     • Implemented various strategies, including a random-coin player and a center-of-mass player.
     • Aimed for a progression in complexity to facilitate performance measurement.

   • Reinforcement Learning:

     • Initial steps involved standardizing the simulation into an "environment."
     • Utilized Proximal Policy Optimization (PPO) as the reinforcement learning algorithm.

3. Training Progress:

   • See videos demonstrating the simulation and hand-coded algorithms.
   • See the win-rate graph of the reinforcement learning algorithm (PPO) against a randomly shooting player.

4. Next Steps:

   • Simulation Enhancement:

     • There is a need for optimization to improve speed and iteration time.
     • Potential adjustments to the physics to better mimic real-life carrom conditions.

   • Hand-Coded Agents Improvement:

     • Could develop more intelligent algorithms with advanced strategies like pocket targeting and defensive maneuvers.

   • Reinforcement Learning Opportunities:

     • PPO algorithm could be optimized.
     • Can try other reinforcement learning algorithms such as Q-learning or model-based approaches.
     • Genetic algorithms like NEAT are potential alternatives for achieving positive results.

Running the environment

1. Ensure that you have Python 3.9 installed.
2. Install the dependencies using python -m pip install -r requirements.txt
3. Run two random agents against each other using python3 main.py

carrom-ai

Gymnasium-based environment

carrom_env

PettingZoo-based environment

auto

Initial translation from https://github.com/samiranrl/Carrom_rl/blob/master/2_player_server/