A hierarchical integrated optimization framework for CAV cooperative merging decision-making and motion control

Human-driven vehicles (HDVs) in mixed on-ramp traffic pose significant challenges to the safety and efficiency of merging manoeuvres for connected automated vehicles (CAVs). This paper proposes a hierarchical integrated optimization framework coupling cooperative merging decision-making with motion control. The upper layer develops a Multi-Agent Reinforcement Learning (MARL)-based cooperative merging model, incorporating a multi-objective reward function and a ramp mapping mechanism to enable adaptive safety decision-making optimization. The lower layer establishes a cooperative motion control model using distributed model predictive control (DMPC) integrated with a priority assignment mechanism to resolve trajectory conflicts effectively. A closed-loop optimization mechanism is formed by feeding control performance back to the upper layer for model refinement and evaluation. Simulation results against internal ablations and external benchmarks demonstrate that the proposed framework improves the merging success rate by more than 7.39% while maintaining superior safety and efficiency across various vehicle counts and CAV penetration scenarios.