Frozen Grid
Agent Navigation Project - CSE 574
As part of Intoduction to Machine Learning , this project showcases an agent's journey through a challenging grid environment. The goal was to develop efficient navigation strategies from start to end.
Environment Overview
The 4x4 grid environment consists of various cell types including frozen, holes, and rewards.
Agent's Objective
The agent aims to maximize rewards while avoiding pitfalls, demonstrating the effectiveness of reinforcement learning.
Technical Details
Implemented SARSA and Double Q-Learning algorithms were pivotal in the agent's decision-making process.
Challenges and Learning
Tuning hyperparameters posed a significant challenge, offering deep insights into the balance between exploration and exploitation.
Conclusion
This project highlights the practical applications of reinforcement learning in navigating complex environments.