Knowledge-Data Dual-Driven Reinforcement Learning for Autonomous Vehicle Control in Mixed Traffic

Summary

In mixed traffic, decision-making for autonomous vehicles faces three interrelated challenges: physics-based priors fail to capture latent interactive intentions and diverse driver behaviors; abrupt maneuvers cause non-stationarity, leaving long-tail safety events under-explored; and hybrid action spaces destabilize unified RL training due to temporal scale differences between continuous car-following and discrete lane-changing. To address these, we propose Knowledge-Data Dual-driven Reinforcement Learning (KDDRL). A conditional deep generative model synthesizes intention-aware future trajectories; a knowledge-data dual-driven paradigm fuses probabilistic data-driven insights with physical constraints to guide safe exploration; and a coupling module compresses intention-aware trajectories and physical constraints into compact shared embeddings, enabling asynchronous multi-timescale optimization. Evaluations on dataset-calibrated simulations show KDDRL effectively handles intention uncertainty, accelerates training convergence, and outperforms baselines in safety, efficiency, and comfort.