PREDICTOR: A tool to predict the timing of the take-over response process in semi-automated driving

Abstract

This paper presents PREDICTOR (PREDICting Take-Over Response time): an interactive open-source research software tool to predict the timing of various stages of a transition of control, or take-over, in semi-automated driving. Although previous work has investigated extensively what factors affect the minimum time needed for a successful take-over by the driver, less is known about how specific stages within the take-over process are affected by those factors. PREDICTOR applies a theoretical framework that describes the take-over process as interruption handling through a series of stages. It then ties this theory to a database that summarizes results from previous take-over studies. PREDICTOR can be used to interactively predict through simulation how specific human factors (e.g., alert modality, alert onset time) impact four distinct stages of the take-over response process. The tool simulates and visualizes expected reaction time distributions for each stage of the take-over process. The use of distributions also highlights the likelihood of an accident – as long responses (“outliers”) are quantifiable. Moreover, it can help understand at which stage drivers might take relatively longer or shorter, and which stages are most impacted by a specific factor (e.g., alert modality). PREDICTOR also allows users to add their own data, and to define their own dependent variables for analysis. As a tool that allows exploration of various scenarios, PREDICTOR can aid in the prediction and analysis of potential future accidents.