关键词: 无人平台, 自动驾驶, 马尔可夫过程, 深度强化学习

Abstract: To address the safety issue of unpredictable behaviors in traditional Deep Reinforcement Learning (DRL) during inference, this paper proposes a safety-enhanced DRL algorithm for autonomous driving in unmanned platforms across multi-task scenarios. The algorithm integrates an improved Markov process with an action evaluation network for pre-execution safety assessment and adopts a parallel dual-thread network architecture to suppress hazardous driving behaviors. Additionally, a novel kinematics-based reward function is designed to balance driving safety and efficiency. Experiments conducted in the highway-env environment across three typical driving scenarios—single-lane roads, intersections, and roundabouts—demonstrate that the proposed algorithm significantly improves driving safety and generalization capability. The results verify its effectiveness and potential for supporting unmanned platform applications such as remote deployment, cargo transportation, and regional penetration.

Key words: unmanned platform, autonomous driving, Markov process, deep reinforcement learning