Reinforcement Learning-Based Adaptive Path Planning for Autonomous Vehicles in Unpredictable Urban Environments

Abstract

Autonomous vehicles (AVs) are transforming transportation by enabling intelligent, self-governing mobility. A crucial challenge in AV deployment is real-time path planning under dynamic and unpredictable road conditions. Traditional algorithms such as A* and Dijkstra, while effective in static environments, lack the flexibility to adapt to unforeseen changes in traffic, road topology, and pedestrian behavior. This paper proposes an adaptive path planning framework using Proximal Policy Optimization (PPO), a state-of-the-art deep reinforcement learning algorithm, integrated into the CARLA simulation environment. The system leverages sensor emulation (LiDAR, camera, GPS) to perceive the environment and trains an autonomous agent to make path decisions that maximize safety, comfort, and efficiency. A customized reward function encourages lane adherence, collision avoidance, and smooth navigation. The trained model is evaluated against classical planning baselines in diverse urban driving scenarios. Experimental results demonstrate improved trajectory stability, reduced collision rates, and higher cumulative rewards, validating the proposed method’s effectiveness in navigating complex, real-time environments. This study serves as a foundational step toward generalizable, intelligent navigation systems in autonomous vehicles.