"But what if it was made of sand..."

Introducing Sandtris, a physics-based twist on Tetris where blocks dissolve into sand and follow gravity upon collision!

Check out the development and results in this YouTube Video!

• Classic Tetris gameplay... but with sand physics!
• Lines are cleared by creating a continuous span of a color of sand from wall to wall.
• Scoring system that rewards more difficult clears.

To get started, you'll need Python 3.

1. Clone the repository:

   git clone https://github.com/pleth9/Sandtris.git

2. Install the required packages using pip:

   pip install pygame torch numpy scipy

To play the game yourself, run the player.py script from the player directory:

python player/player.py

To train the AI model, run the train.py script from the AI directory:

python AI/train.py

Note: Training is computationally intensive. The code is configured for Apple's MPS for Apple Silicon Macs. Use CUDA for NVIDIA GPUs or just CPU backend.

The current AI is a Convolutional Neural Network (CNN) trained with Deep Q-Learning (DQN). The model (tetris_nn.py) and training script (train.py) are located in the AI folder.

As of 8/13/25, I have paused development and training, as I need my laptop back for work reasons. I encourage you to experiment with the reward function, model architecture, and training process!

Please feel free to make a PR with your updated training scripts or saved brains! I am greatly curious as to how the reward function can be improved alongside other parts of the training process!