Towards Enhancing Driver’s Perceived Safety in Autonomous Driving: A Shield-Based Approach

This paper proposes Personalized Perceived Safety Shielding (PPSS), designed to enhance perceived safety in reinforcement learning applications for autonomous driving. Utilizing a shield synthesis strategy, the framework ensures alignment with the driver’s perceived safety preferences, derived from human driving data. Our framework intricately considers drivers’ safety preferences, using real-world data for a personalized shielding strategy. This personalizing aligns with the human element in driving, addressing diverse safety perceptions effectively. The paper evaluates the proposed PPSS in terms of reflecting personal safety preferences and demonstrates its impact on the objective safety of driving. Through a detailed analysis of obstacle distances, the trajectories associated with the shield distance are examined. Preliminary results indicate that the trajectories generated by our framework are not only compatible with drivers’ perceived safety needs but also maintain and enhance objective safety measures.