Validación de una estrategia para la estimación del riesgo en intersecciones con vehículos conectados

To make possible the massive deployment of automated vehicles in urban environments, it is essential to move forward in making safe decisions. In particular, it is necessary to improve the ability to infer the intentions of the di erent agents and the risk involved in complex driving scenes, thus improving the safety and predictability of automated driving and driving assistance systems. The present work shows the implementation and validation in simulation of a novel solution to estimate the risk of driving using a model of space and states in the context of intersections. The strategy used models the driving scene as a dynamic Bayesian network and infers intentions and expectations of the agents involved through a particle filter. The results are very promising in both the success rate and the prediction horizon in the environments for which it has been tested: Y, T and X intersections.