Snake with Curriculum Learning
Training an RL agent to play Snake using Stable Baselines3, with curriculum learning to progressively increase difficulty.
This project implements a reinforcement learning agent for the classic Snake game using Stable Baselines3.
A curriculum learning strategy is applied: the agent is trained on smaller, simpler boards before progressing to larger and more complex environments.
Methods
- Custom Snake environments in snakeenv.py and snakeenv2.py
- RL training with Proximal Policy Optimization (PPO)
- Curriculum schedule: gradually increasing board size and difficulty
- Logging and monitoring with Stable Baselines3 tools
Key results
- Agent successfully learned to play Snake across multiple board sizes
- Achieved a maximum score of 18 apples eaten in evaluation runs
- Curriculum learning improved convergence speed and stability compared to direct training on large boards
Reinforcement learning agent trained with curriculum learning to master Snake