This study explores the application of four advanced reinforcement learning (RL) algorithms—Proximal Policy Optimization (PPO), Deep Q-Networks (DQN), Advantage Actor-Critic (A2C), and Deep Deterministic Policy Gradient (DDPG)—to optimize agent performance in the Doom environment, a high-dimensional first-person shooter game. Each algorithm was trained to tackle tasks such as navigation, target elimination, and resource collection under varying levels of complexity. The experiments incorporated custom reward engineering, state- space design, and environment modifications to challenge agent adaptability. Performance evaluations revealed that PPO achieved superior stability and sample efficiency in continuous action scenarios, while DQN excelled in discrete state-action spaces with deterministic rewards. Recent applications have demonstrated the potential of reinforcement learning in controlling game agents to achieve victory and its role in facilitating game development processes for human developers. A2C demonstrated strong scalability and rapid convergence by using synchronous updates with both actor and critic networks. DDPG showcased proficiency in learning continuous control policies but required significant hyperparameter tuning for optimal performance. Across all experiments, PPO and A2C exhibited the most consistent performance in handling complex game mechanics. Comparative analysis of training time, computational resource usage, and convergence rates highlighted the trade-offs inherent in each approach. The findings underscore the efficacy of RL techniques in mastering dynamic, interactive environments like Doom and provide insights into their practical deployment for real-world applications such as robotics, autonomous navigation, and strategic decision-making systems.