Abstract

An important part of Advanced Planning Systems are algorithms. When algorithms are applied, the task is automated as much as possible. However, the human that is supposed to use the algorithm is generally ignored during the development process. As a consequence, a prior investigation whether and how an algorithm can or will be used in practice is not integrated in the development process. In contrast, in the field of cognitive ergonomics, function allocation methods explicitly take into account human factors in the design of human/computer systems. The function allocation literature, however, is mainly focused on dynamic systems where humans must make decisions in situations with time pressure and important safety risks, e.g., nuclear plants and air traffic control. We analyze the differences between such dynamic systems and planning and scheduling, and we propose a model for function allocation in planning and scheduling taking into account cognitive and human-machine cooperation aspects.

Abstract

International audience; —This paper presents a study carried out within the context of Level 3 automated driving. It aims to characterize drivers takeover behaviour by exploring quantitative data related to the driver's performance (vehicle data) and qualitative data (user experience expressed in post-activity interviews). For this purpose, several techniques of multivariate analysis were used such as clustering of variables and units. This study illustrates that user experience is associated with the magnitude of actions related to lateral and longitudinal control; positive experience being associated with smoother actions, whereas negative feelings were associated with rougher ones. Furthermore, two main driver's profiles emerged; they are described by two latent variables, one of which is quality of the actions. Future research will investigate the determinants of these profiles.

Abstract

International audience; This text aims to present a study which deals with mental workload evaluation during task execution. It is focused on the Air Traffic Controllers working situation. In this document, we mainly introduce an experiment which has been conducted in a French En-Route air traffic center with the participation of Air Traffic Controllers. Four principal experiment characteristics are detailed: the experiment procedure, the working situation elaborated for our experimentation, the nature of the task achieved by participants, and the technique chosen to analyze mental workload felt by operators. We finally present the main results from our first data analysis which seem to confirm major observations known in the field of air traffic control, as well as, mental workload study field.

Abstract

International audience; One way to improve both capacity and safety in air traffic control (ATC) implies to assist the human operator, while he remains in the decision loop. Designing high performance tools requires a precise tuning in order to interface them efficiently and non-intrusively with the controller, especially in heavy workload contexts. To this end, knowledge of workload sources and fluctuations is essential and has triggered a large number of studies. Surprisingly, there is no universally accepted definition of mental workload: a consensus exists merely to define it as the "cost" of a given task for the operator. Concept of mental load is related to information processing theory Limited capacity of cognitive resources and sequential processing of information define what can be overload. an excessive demand on perceptual and cognitive resources (visual and auditory perception, memory and attention, among others) with respect to information processing capacities. In the present study, we would like to propose a different way to model the link between aircraft configuration and perceived workload.

Abstract

International audience; In this paper, we compare several machine learning methods on the problem of learning a model of the air traffic controller workload from historical data. This data is a collection of workload mesurements extracted from past sector operations and of ATC complexity measurements computed from radar records and airspace data (sector geometry). We assume that the workload is low when a given sector is collapsed with other sectors into a larger sector, normal when it is operated as is, and high when it is split into smaller sectors assigned to several working positions. This learning problem is modeled as a classification problem where the target variable is a workload category (low, normal, high) and the explanatory variables are the air traffic control (ATC) complexity metrics. Several classifiers are compared on this problem: linear dis-criminant analysis, quadratic discriminant analysis, naive Bayes classifiers, neural networks, and gradient boosted trees. The performance of these models is assessed on a separate test set. The best methods show a rate of correct predictions around 82%.