International Choice Modelling Conference, International Choice Modelling Conference 2019

Font Size: 
Deriving metrics of driving comfort for autonomous vehicles. A time-series latent variable approach of speed choice.
Evangelos Paschalidis, Foroogh Hajiseyedjavadi, Chongfeng Wei, Albert Solernou Crusat, Natasha Merat, Richard Romano, Erwin Boer

Last modified: 15 July 2019

Abstract


Over the past decade, technological advances and developments of autonomous vehicles (AVs) have been among the spotlight topics for the automotive industry and transportation research community. The gradual substitution of manually driven vehicles by AVs is expected to have substantial impacts towards the improvement of road safety and the reduction of congestion and fuel consumption. However, the degree of AV integration in the future transportation systems highly depends on user acceptance and thus intention to use. These issues range from willingness to purchase, to safety perception and morality issues. From a passenger’s perspective, intention to use AVs has been approximated as a function of attitudes and psychosocial factors [1] with a different stream of research, in the fields of human factors and engineering, focusing on the impacts of perceived comfort and safety that emerge from the performance of AVs. However, existing literature lacks a universally established definition of comfort while navigation and performance of these vehicles is currently performed via the use of environmental and infrastructure-based cues with sensors used for obstacle detection and avoidance [2]. To that end, drivers’ perception about AV performance, has been mostly based on metrics such as appropriate speed, lateral behaviour and acceleration [3] while other studies [4] have attempted to derive similar metrics of comfort and safety via manual driving observations.

The aim of this paper is to focus on modelling drivers’ speeding behaviour, as a function of the of the road environment. The analysis is based on data collected as a part of the UK-funded HumanDrive project, the main purpose of which is the development of human-like controls for AVs. The data collection was conducted at the University of Leeds Driving Simulator, where participants drove four simulated scenarios of varying risk composition. Each scenario consisted of several 250m long segments and the variability in risk was applied through changes in the road type (urban/rural), lane width, curvature, contextual lateral risk (e.g. hard, soft, and raised roadside, parked cars etc.), oncoming traffic and persistence of risks along segments. On top of the driving tasks, participants also completed a series of questionnaires (locus of control, sensation seeking etc.). In previous analysis of the data, significant effects of the risk levels were found on longitudinal and lateral vehicle control [5], while speeding behaviour was found to correlate with participants’ sensation-seeking [2] as it was derived from the Arnett Inventory of Sensation Seeking items [6].

This paper builds on the aforementioned data analyses and suggests a modelling framework to combine the effects of road environment and drivers’ characteristics on speed choice. The latter has been averaged across individuals for each 250m segment, and treated as a dependent variable in a time-series model, where the road characteristics have been used as explanatory variables. As an individual’s speed choice is expected to be influenced by past behaviour, speed from the previous time step (road segment) has been lagged and included as an independent variable. Additionally, the model accounts for unobserved heterogeneity, using a standard normal disturbance component. To address the potential correlation between the disturbance term and the lagged speed dependent variable, the approach suggested by Wooldridge [7] has been applied and the model has been estimated conditionally on the initial observation of each individual. Moreover, an additional correlation parameter has been introduced to incorporate disturbances that follow an autoregressive process, similar to the specification for maximum likelihood estimation described in [8]. Finally, sensation-seeking has been considered as a latent variable using indicators and an ordinal variable specification has been used to define their likelihood.

The main findings show that speed decreases significantly as road radius reduces, while the same pattern is seen for lane width. Moreover, the lateral risk level producing the highest negative impact on speed is the presence of parked cars on the road. Also, past speeding behaviour has a positive effect. Finally, sensation-seeking was also found to significantly contribute to the speed increase. In summary, the findings denote that the road environment has a major impact on speed choice, however there are more aspects to be considered. For instance, the significant effect of sensation-seeking implies that different drivers may have different expectations about the most preferred driving style of an AV controller. Also, the significant effect of the random disturbance term consists an indication that speed choice is a complex issue that varies across individuals. Based on the current findings, there is scope for further research to investigate additional metrics of longitudinal and lateral driving performance, that could lead to the determination of their acceptable boundaries and thus the development of more human-like AV controllers that will positively contribute to the acceptance of this technology.

References:

  1. Buckley, L., Kaye, S. A., & Pradhan, A. K. (2018). Psychosocial factors associated with intended use of automated vehicles: A simulated driving study. Accident Analysis & Prevention115, 202-208.
  2. Louw, T., Hajiseyedjavadi, F., Jamson, H., Romano, R., Boer, E., & Merat, N. Relationship between sensation seeking and speeding behaviour in road environments with different contextual risks. (Under review)
  3. Hartwich, F., Beggiato, M., & Krems, J. F. (2018). Driving comfort, enjoyment and acceptance of automated driving–effects of drivers’ age and driving style familiarity. Ergonomics, 1-16.
  4. Bellem, H., Schönenberg, T., Krems, J. F., & Schrauf, M. (2016). Objective metrics of comfort: developing a driving style for highly automated vehicles. Transportation research part F: traffic psychology and behaviour41, 45-54.
  5. Hajiseyedjavadi, F., Louw, T., Jamson, H., Merat, N., Boer, E., & Romano, R. (2019) Road factors and their impact on drivers’ speed and lateral position control behaviour. Accepted at: The 10th International Driving Symposium on Human Factors in Driver Assessment, Training, and Vehicle Design, June 24-27, 2019 Santa Fe, New Mexico
  6. Arnett, J. (1994). Sensation seeking: A new conceptualization and a new scale. Personality and individual differences16, 289-289.
  7. Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press.
  8. Davidson, R., & MacKinnon, J. G. (2004). Econometric theory and methods (Vol. 5). New York: Oxford University Press.

 


Conference registration is required in order to view papers.