A Feasibility Study of Risk Triggering Using Multimodal Trajectory Predictions in Mixed Traffic

Abstract

Ensuring safe interactions between vehicles and vulnerable road users (VRUs) is a critical challenge in mixed traffic environments. This study investigates the use of multimodal trajectory predictions to anticipate potential hazards involving VRUs and generate actionable risk warnings. A structured framework is proposed to extract safety indicators from predicted trajectories and evaluate multiple trigger strategies. Experimental results show that OR-logic aggregation achieves the highest overall performance by balancing hazard detection and false alarm control, while other strategies exhibit more conservative or imbalanced behavior. Furthermore, the choice of the number of aggregated trajectories (K) introduces an additional trade-off, where increasing K enhances risk coverage but may lead to reduced reliability due to more frequent warnings. These findings provide practical guidance for the design and deployment of risk warning systems.