کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

Lane Change Decision Making Using Deep Reinforcement Learning

نویسندگان :

Pedram Lamei¹ Mohammad Haeri²

1- Sharif university of technology 2- Sharif university of technology

کلمات کلیدی :

Deep reinforcement learning،Decision making،Lane change،Autonomous vehicles،Autonomous driving

چکیده :

This paper explores the application of deep reinforcement learning for decision-making in autonomous vehicle lane-changing scenarios. Lane changes, a critical aspect of driving, pose significant challenges for autonomous systems due to complex traffic dynamics and safety constraints. The study investigates two deep reinforcement learning algorithms, deep Q-network and proximal policy optimization to train agents for efficient and safe lane changes in a stochastic highway environment. By defining a structured state, action, and reward space, the proposed methods emphasize collision avoidance, adherence to traffic norms, and travel efficiency. Simulation results reveal distinct strengths of two approaches. Deep Q-network demonstrates aggressive efficiency with higher overtakes and rapid transitions, while proximal policy optimization prioritizes safety through conservative strategies, achieving consistent maximum distances. Comparative analysis highlights the trade-offs between these methods, offering insights for developing robust autonomous driving policies. This research contributes to advancing intelligent transportation systems by addressing decision-making challenges and promoting adaptive learning for enhanced vehicle autonomy.