Driver Activity Recognition by Means of Temporal HTN Planning

Juan Fernandez-Olivares, Raul Perez

 PosterID: 22
picture_as_pdf PDF
library_books Poster
menu_book BibTeX
Aimed to increase road safety and improve transport drivers' working conditions, authorities world wide have imposed complex hours of service (HoS) regulations which constraint the limits on the amount of drivers' working and driving time without resting. Therefore scheduled workplans have to be aligned with laws that define the legal behavior of a driver, what complicates the way in which they have to be enacted. At present, every company's transport assets are endowed with electronic devices (tachografs) which record the different activities a driver performs when delivering a transport service. This enables drivers workplan schedules to be monitored and recorded either by authorities and transport companies. Therefore, both drivers and companies are interested not only on driver’s compliance of regulation, but also on identifying whether the driver’s style of driving corresponds with any of the behaviour patterns defined by the HoS regulation. This enables to know ahead of time whether a service can be effectively delivered or the driver behavior may be subject to penalty, which can have a very negative impact on the costs and profitability of a service. In any case, recognizing driver's driving activity is a very relevant problem for transport companies. In this paper we present an application that, starting from a real event log extracted from a tacograph, identifies different subsequences of the driver's driving activity and labels them according to the terms defined by the HoS regulation. The identification of the temporal subsequences is an activity recognition problem that is addressed as a temporal HTN planning problem wherein the domain describes the HoS regulation constraints as a hierarchy of tasks with temporal and numerical constraints, the initial state represents a set of temporally annotated observations each one corresponding to an event of the original log, and the goal-task can be seen as a parsing task to be performed considering the temporal HTN domain as a set of production rules of an attribute grammar. The result of this recognition process is a labeled event log that can be easily interpreted by company experts which can made more informed decisions considering the historic or current labeled situation of a driver. We have implemented a proof of concept and carried out an experimentation in order to validate the application by successfully identifying subsequences in a dataset representing a real event log with more than 40K events.

Session E4: Applications
Canb 10/28/2020, 01:00 – 02:00
10/30/2020, 18:00 – 19:00
Paris 10/27/2020, 15:00 – 16:00
10/30/2020, 08:00 – 09:00
NYC 10/27/2020, 10:00 – 11:00
10/30/2020, 03:00 – 04:00
LA 10/27/2020, 07:00 – 08:00
10/30/2020, 00:00 – 01:00
Contact: icaps20-support (at) googlegroups.com
Copyright ©2020 by ICAPS.