Multi-modal intelligent driving decision-making in complex traffic scenarios integrating DDQN and DDPG

In complex traffic environments, the interaction between vehicles is highly dynamic and significantly nonlinear, which poses a challenge to the coordinated optimization of safety and stability in intelligent driving decision-making. To enhance adaptability and decision-making accuracy in complex scenarios, a multi-modal intelligent driving decision model integrating continuous control and discrete decision-making is developed. The model is based on a longitudinal continuous control model with deep deterministic policy gradient to achieve adaptive acceleration adjustment, and introduces a multi strategy particle swarm optimization mechanism to improve the convergence stability of the strategy. On this basis, a hierarchical decision-making structure is constructed by combining dual deep reinforcement learning methods, where high-level discrete behavior decisions and low-level continuous motion control are collaboratively optimized through temporal modeling and priority experience learning. The research results show that in terms of car following control, the average distance between vehicles reaches 20.1 m, which is about 8.1% higher than the comparison model. The collision rate drops to 1.3%, which is more than 55% lower than the benchmark model. The average headway is 1.95s, with significantly reduced fluctuations. The average acceleration decreases to 0.72 m/s 3 , a decrease of about 23%. The comfort satisfaction rate increases to 93.8%, and the safety constraint satisfaction rate has reached 96.5%. The research shows that the constructed model achieves significant improvements in safety, stability, comfort, and real-time performance, which can provide effective support for intelligent driving in complex traffic scenarios.