Intelligent defensive driving for autonomous vehicles: Framework, strategy and verification.

As autonomous driving advances to higher levels, conventional decision-making algorithms for autonomous vehicles (AVs) often inadequately address long-tail issues composed of low-frequency, high-uncertainty, and extreme scenarios, leading to recurrent safety incidents. How to leverage humans' excellent defensive driving experience refined through expert knowledge to upgrade autonomous driving to intelligent defensive driving, thereby systematically improving driving safety serves as the core objective of this paper. To tackle this challenge, this paper proposes an integrated research scheme for the intelligent defensive driving of AVs. Firstly, an overall framework for intelligent defensive driving of AVs is constructed, which includes a systematic classification method for defensive driving scenarios and a hierarchical design process for defensive driving. Secondly, an online defensive driving monitoring mechanism for AVs is designed based on experience-triggered conditions, and a safety decision-making strategy is developed integrating formalized defensive driving experience. Finally, intelligent defensive driving performance for AVs is verified based on the "perception insufficiency" classification. The simulation results show that the proposed research scheme can utilize the defensive driving trigger mechanism to anticipate potential risks in multiple risk scenarios, improving the average values of the minimum relative braking distance, minimum time-to-collision and advance braking time by 7.49 m, 1.56 s, and 2.84 s, respectively, while significantly reducing the probability of emergency accidents and maintaining system robustness and real-time computational performance.