Deriving metrics of driving comfort for autonomous vehicles: A dynamic latent variable model of speed choice

Abstract While the interest of the transport research community and automotive industry is increasingly turning towards developments and improvements in the field of autonomous vehicles, there is a need for a better understanding of the end users’ preferences regarding perceived passenger comfort, in order to improve acceptance and intention to use. The present study is based on a driving simulator experiment conducted at the University of Leeds Driving Simulator and approaches the issue of comfort via observed speed choice behaviour. Participants drove a series of driving simulator scenarios composed of road segments of different road type, road geometry, risk level at the road edge, and oncoming traffic. They also completed a series of self-report questionnaires, including Arnett’s Inventory of Sensation-seeking. A set of models was developed in order to investigate the effects of road environment and sensation-seeking on speed behaviour. The initial model only considered explanatory variables related to the road environment and accounted for individual unobserved heterogeneity. Past behaviour, serial correlation and heterogeneity in road environment were then introduced in the model specification. The autoregressive disturbance term that accounted for serial correlation was also applied in the form of a random variable and significantly improved model fit. Finally, sensation-seeking was incorporated in the model as a latent variable. The results showed a significant impact of most of the road elements as road type, curvature, risk type at the road edge on observed behaviour, implying a future need for the development of autonomous vehicle controllers that adapt their performance based on the road environment. Moreover, sensation-seeking had a significant and positive effect on speed, which indicates a potential future demand for personalised controllers to meet the users’ individual preferences.