Training an AI-Powered Doomguy Leveraging Deep Reinforcement Learning
Boyi Xiao · Advances in intelligent systems research/Advances in Intelligent Systems Research · 2024
Reinforcement learning (RL) has recently gained significant attention due to the impressive successes of self-driving cars and human-like performance in games such as Go or StarCraft.However, approaching this subject can be intimidating.In this research, the author aims to explore how to train a RL agent to play various Doom scenarios.This will provide an opportunity to explore various aspects of RL, such as curriculum learning, reward shaping, and machine learning in general.The author will also address how to monitor the agent's progress during training and how to fix any issues that arise.Monitoring is crucial to ensure that the agent is learning effectively and behavior is appropriate.If any issues are detected, they can be fixed by adjusting the training process or the reward structure.The ultimate goal of this research is to train an RL agent to play deathmatch against real human players, with the author replacing humans with ingame bots.At the end of this article, you will see a human-like agent playing against bots like a real player.The author hopes to demonstrate the potential of RL in creating intelligent and autonomous agents that can compete against humans in complex environments.