Inspired by Sandtris—a physics-based Tetris game where blocks break into sand—this project combines my interests in machine learning, gaming, and physics. The goal is to create a reinforcement learning (RL) model that learns to play and maximize its score within this dynamic environment.
Attending the NYU AI School Summer Program sparked my interest in reinforcement learning, but I wanted a deeper dive. This project serves as a hands-on opportunity to explore RL and create a project that I can showcase to professors and future employers.
Training an AI for a game with complex, particle-based physics is challenging. The model must not only understand game rules but also adapt to the unique physics governing each block's movement. The goal is to maximize the game score by navigating these dynamics effectively.
- RL Model: Built with TensorFlow or PyTorch
- Game Environment: Using an open-source Sandtris game
- Training Setup: Local training with progressive optimization
- Documentation: Slides and video demos showcasing results
- Framework: TensorFlow / PyTorch
- Environment: Open-source Sand Tetris game
- Training: Local machine
- Documentation: Slides and video demos
- Set up game environment
- Implement RL model
- Conduct initial training
- Optimize and fine-tune
- Document progress and create demo video
This project will deepen my understanding of RL within a real-time, physics-based setting and enhance my technical skills. I'm excited to push this model's performance and share updates as I refine it.