This project features a custom-built Reinforcement Learning (RL) environment named Windy Drone, created from scratch. The environment is a 6×6 gridworld where a drone must navigate from a random starting point to a random destination while avoiding obstacles and managing its battery and health.
It includes implementations of several RL algorithms such as Q-learning, SARSA, n-step Tree Backup, and Monte Carlo which also developed from scratch. Each algorithm is thoroughly tested in this environment, with detailed pseudocode, implementation examples, and performance results provided. The environment’s reward system is influenced by Gaussian distributions, and episodes can be truncated based on battery and health levels.
For more details, read this.
This project includes an implementation of Q-learning developed from scratch. It is thoroughly tested in the Frozen Lake environment, and detailed pseudocode, implementation examples, and performance results are provided.
For more details, read this.