Reinforcement Learning-Based Pathfinding For Autonomous Agents

Abstract

This thesis investigates the use of Reinforcement Learning (RL) algorithms, specifically QLearning and SARSA, as potential pathfinding solutions in a Unity 2D game environment. This research explores the capabilities of RL as an alternative approach for autonomous agents within a gaming framework. The implementation involves creating a training scene where a single monster agent learns to navigate the game environment. The agent receives rewards for actions that bring it closer to the player and penalties for actions leading to less desirable outcomes. Through training, both the Q-Learning and SARSA models adjust their policies based on these rewards and penalties, leading to improved pathfinding behavior over time. The thesis provides a comparative analysis of Q-Learning and SARSA, examining their effectiveness and efficiency in pathfinding tasks. The findings indicate that reinforcement learning holds significant potential as an adaptive alternative in complex scenarios. However, further research and development are necessary to fully realize this potential. Future work could explore more complex environments, multiple agents, and advanced RL techniques to enhance the capabilities and applications of this approach. This study provides valuable insights into using RL for game AI, highlighting both opportunities and current limitations.