Exploración de Redes Neuronales Holográficas con Cuantificación Difusa para el Monitoreo de Conductores en Conducción Autónoma Condicional

The Society of Automotive Engineers (SAE) defines Automated Driving Systems (ADS) for road vehicles as being that can perform the entire dynamic driving task without a human driver in the loop. Under conditional driving automation (SAE Level 3), when automated driving fails the drivers are expected to resume manual driving. For this transition to occur safely, it is imperative that drivers react in an appropriate and timely manner, which is difficult to happen once the driver has been subjected to long distances of autonomous driving. Artificial Intelligence (AI) techniques could be used for safety assurance of adaptive safety-critical systems. Not only sensing the external environment of the vehicle, but also monitoring the state of the driver-vehicle communication. Further, the concept of explainable AI was highlighted as having potential to provide evidence from ADS that could support safety assurance and regulatory compliance. In this work we present a neuro-fuzzy method working as an explainable machine learning approach suitable for domains where validation of the underlying non-linear prediction models is required. The results of comparison between proposed model and other models from the literature show that the proposed model could provide explanations about its predictions in real time to ensure smooth transitions in SAE Level 3