Two-stage remote takeover strategy for autonomous driving based on driving risk fields

Addressing safety issues arising from safety officers failing to take over in a timely manner and inadequate risk warnings during remote takeover in autonomous driving, this paper focuses on the quantification of driving risks and takeover control strategies. First, a unified model for driving risk fields is established. By calibrating equivalent mass and environmental condition influence coefficients, the model comprehensively quantifies static entities, dynamic entities, and environmental factors, defining the forces acting on the vehicle as driving risks. Second, based on this risk model, a two-stage takeover model is designed. Risk sudden changes are identified by the average value and standard deviation of risk within a time window. Warning and takeover request strategies are formulated based on the driving risk change rate and the relative driving risk safety index. A hierarchical warning mechanism is constructed using a visual-auditory multimodal approach to enhance the safety officer's response efficiency. Finally, in-the-loop simulation verification is conducted in scenarios involving fog, malfunctioning vehicles, and sudden traffic accidents. The results show that the proposed strategy can trigger warnings an average of 2 seconds earlier. In traffic accident scenarios, successful takeover reduces risk by 51.35% compared to non-takeover scenarios, validating the effectiveness and reliability of the strategy in reducing untimely takeover situations and enhancing takeover safety.