ARCHIVES

This study explores how reinforcement learning (RL) algorithms like PPO and DDPG can train autonomous systems in simulated environments such as CARLA. We designed driving tasks like lane following and obstacle avoidance, evaluated performance using success rate and collision metrics, and compared RL agents with traditional controllers. The results show that RL methods learn effective driving behaviors over time. Our work highlights key contributions in simulation-based RL training and identifies limitations such as sensitivity to environment changes. Future research will focus on real-world transfer, multi-agent coordination, and sim-to-real adaptation.