Safety assurance of intelligent driving: system performance definition

As technology has always followed an upward pattern, concepts that were once the preserve of science fiction are now manifesting themselves in reality. One once-fictional concept now emerging in the automotive industry is autonomous driving. Autonomous driving systems in vehicles are often referred to by one of several interchangeable names, both in science fiction and in modern colloquial language, including "self-driving cars," "driverless cars," or "self-driving cars." In recent years, the use of these terms has surged as numerous industry leaders and pioneers explore the possibility of realistically implementing them in the near future. However, these efforts have found uncertainty. Companies are struggling to figure out where to start and what factors to consider. Fortunately, standardization has long helped members of any industry move forward with minimal risk. For driverless cars, standardization makes the concept a much easier concept to address.

An important factor guiding the automotive industry as it develops, tests and ultimately mass-produces self-driving cars is the terminology surrounding the concept, which can help shape the industry's path forward.

By proposing taxonomy and outlining generally accepted definitions around driving automation, SAE J 3016-2021, with its descriptive and informative format (vs. prescriptive), provides a welcome clarity that successfully stabilizes the topic of driving automation and saves considerable time and effort. In addition, SAE J 3016-2021 clarifies the role of the human driver in the engagement of driving automation systems, provides a useful framework for the specification of driving automation, and answers questions and concerns that may aid in the development of laws, policies, regulations, and standards. Specifically, SAE J 3016-2021 describes motor vehicle driving automation systems that continuously perform some or all of the dynamic driving task (DDT) and introduces the three main roles in driving: human users, driving automation systems, and other vehicle systems and components. Autonomous Vehicles? The wording around automotive driving systems helps to best position the automotive industry. For example, SAE J 3016-2021 does not use terms such as "autonomous vehicle" or "driverless car." In fact, its wording is careful to refer to automation as part of a vehicle system, rather than automation in the vehicle itself. The classification refers to automated functions, of which there may be several in a single vehicle. SAE J 3016-2021 identifies "autonomous vehicle" as obsolete because, while some vehicles may be highly independent and self-sufficient, their functionality will rely on communication and cooperation with external entities. For these same reasons, "robotic vehicle" is misleading, while "robot" is a vague term often attached to advanced technology. As for "autonomous driving," "unmanned driving," or "driverless driving," there is a clear misunderstanding of the different levels of SAE through their use. "Driver" is a term that can mean anything and is not necessarily specific to the human individual turning the steering wheel. As such, these terms confuse rather than clarify. Ultimately, the official correct term to refer to vehicles with these systems is "motor vehicle with an automated driving system," according to SAE J 3016-2021.

In conditional automated driving, drivers have difficulty taking over control when requested. To address this challenge, this paper aims to predict the driver's takeover performance before a takeover request (TOR) is issued by analyzing the driver's physiological data and external environment data. We used a dataset of two-person in-the-loop experiments, in which drivers engaged in non-driving related tasks (NDRTs) were asked to take over control of the automated driving in various situations. The driver's physiological data included heart rate index, skin electrochemical response index, and eye tracking metrics. The driving environment data included scene type, traffic density, and TOR preparation time. The driver's takeover performance was classified as good or bad based on their driving behavior during the transition period and was taken as the ground truth. Using six machine learning methods, the random forest classifier was found to perform best and was able to predict the driver's takeover performance when performing NDRT under different cognitive load levels. Three seconds was suggested as the optimal time window for predicting the takeover performance using the random forest classifier, with an accuracy of 84.3% and an F1 score of 64.0%. The results are meaningful for the algorithm development of driver state detection and the design of adaptive in-vehicle warning systems in conditional automated driving.