Probabilistic Risk Metric for Highway Driving Leveraging Multi-Modal Trajectory Predictions

Road traffic safety has attracted increasing research attention, in particular the current transition from human-driven vehicles to autonomous vehicles. Surrogate measures of are widely used assess but they typically ignore motion uncertainties and inflexible dealing with two-dimensional motion. Meanwhile, learning-based lane-change trajectory prediction models have shown potential provide accurate results. We therefore propose a prediction-based driving risk metric for on multi-lane highways, expressed by maximum value over different time instants within horizon. At each instant, vehicle is estimated as sum weighted risks mode finite set maneuver possibilities. Under mode, calculated product three factors: probability, collision probability expected crash severity. The factors leveraging two-stage multi-modal predictions surrounding vehicles: first intention module invoked possibilities, then possibilities partial input module. Working empirical dataset highD simulated highway scenarios, proposed model achieves superior performance compared state-of-the-art model. computationally efficient real-time applications, effective identify crashes earlier thanks employed