An AIoT-Enabled Digital Twin CAVs With a DRL-Based Framework for Trajectory Planning

The convergence of intelligent transportation systems and urban informatics has given rise to the deployment of connected and autonomous vehicles (CAVs) which offers the potential to enhance the safety and efficiency. However, the increasing volume of automobiles on highways causes frequent and often mismanaged multi-lane changing (MLC), coupled with inadequate trajectory planning. This results in traffic congestion and accidents, which leads to substantial societal losses. Additionally, these issues raise substantial concerns about environmental sustainability, safety, and traffic efficiency, necessitating innovative solutions. To address these challenges, we leverage the transformative capabilities of Artificial Intelligence of Things (AIoT) and introduce a deep reinforcement learning (DRL)-based non-cooperative game approach, named Nash-SAC (Soft Actor-Critic), enabled by digital twin technology, to facilitate optimized decision-making in CAVs. We consider various driving behaviors and social interaction characteristics that influence driving safety, ride comfort, and travel efficiency. The efficacy of the proposed framework is validated through simulations using the Python-based Highway-env simulator and Matlab/Simulink. The simulation analysis reveals that the proposed algorithm attains 22.48%, 40.32%, and 52.02% reductions in average delay, and achieves 39.50%, 58%, and 64.46% lesser computational time compared to the Twin-Delayed Deep Deterministic Policy Gradient (TD3), Deep Deterministic Policy Gradient (DDPG), Deep Q-Network (DQN) algorithms, respectively.