Towards Human-in-the-Loop Cyber-Physical Systems

Abstract

Listen

Translate article

The present work tackles four common aspects of the human-in-the-loop challenge, one of the major questions in cyber-physical systems (CPS) related research. First, strategic decision measures are identified in a divided attention task not considered by previous studies, and their importance is shown in predicting human performance. This supports the creation of a more complete human behavior model to be integrated into CPS. Second, a generic data-driven approach is proposed for predicting human errors from eye-gaze and hand-motion features with high accuracy. Anticipating human errors facilitates efficient computer intervention and the reliable operation of complex systems. Third, it is shown that demanding gaze-based control of interfaces can be productive in terms of strategies, even though it impairs performance. This promotes intuitive interaction with computers and is especially important in cases where traditional control methods are not feasible. Fourth, an intuitive monocular vision based ego-speed estimation, and a time-to-collision prediction algorithm is investigated using as input two video streams, recording the frontal road view and the driver's perspective. Leveraging smartglasses as sensory devices and combining them with deep learning algorithms improves the decision making of human assistance systems. The results contribute to increasing human-awareness of CPS and to incorporating humans into the loop as an integral part.