Deep Reinforcement Learning Actor-Critic Algorithm based Autonomous Lane Change Decisions in Static Environments

Deep Reinforcement Learning (DRL) has emerged as a pivotal technology in modern transportation systems, offering considerable potential to enhance driving safety and optimize performance. This study underscores the significance of DRL in autonomous driving (AD) and introduces the Advantage Actor-Critic (A2C) approach for lane change decision-making. A2C, representing policy-based methods, is compared with the Prioritized Replay Deep Q-Network (PRDQN), a value-based approach previously used in similar scenarios. Both approaches were evaluated using the CARLA (Car Learning to Act) simulation environment, focusing on static obstacle scenarios. The comparative analysis aims to assess the performance differences between A2C and PRDQN, highlighting the effectiveness of A2C in improving decision-making accuracy, safety, and adaptability. The results demonstrate the superior performance of A2C in improving safety and decision-making within autonomous driving systems.