Comparative Analysis of Reinforcement Learning Algorithms for Autonomous Driving in Simulated 2D Environments: Optimizing Reward Functions and Hyperparameters

This paper presents a comparative analysis of reinforcement learning (RL) algorithms, specifically Proximal Policy Optimization (PPO) and Deep Q-Network (DQN), for autonomous driving in simulated 2D environments. The study focuses on optimizing reward functions and hyperparameters to enhance road navigation and obstacle avoidance. Our experiments show that DQN generally outperforms PPO in simple environments, while fine-tuning reward structures and hyperparameters significantly impacts the learning process. Techniques such as frame stacking and curriculum learning further improve performance.