ABSTRACT
The aim of this study was to extend data on user-tailorable takeover interfaces for semi-autonomous cars and to identify clusters within drivers’ customisation profiles. These clusters shall help future machine-learning techniques to improve performance. High inter-driver variability is the main motivation for user-tailorable interfaces which have been applied in the automotive sectors for decades. However, there is little data on user-tailorable interfaces for control transitions and scarce evidence regarding their benefits. In a naturalistic on-road study on the UK motorway, 24 participants were exposed to a user-tailorable takeover interface. Over two trials, they experienced six takeover requests in total, each following either 8 or 10 min of automated driving. In the first trial, participants experienced a default takeover interface. In the second trial, they experienced a tailored interface which they had customised beforehand. At the end of trial 2, participants were allowed to make final adjustments. The results showed that 20 out of 24 customisation profiles were unique, highlighting the individuality of each driver. The hierarchical agglomerative cluster analysis revealed two clusters, which differed in their usage of the infotainment display, indicating high inter-driver variability. Significant differences between the identified clusters in this on-road study and the ones from a previous simulator study demonstrated a high intra-driver variability; drivers from the same cluster adapted differently to environmental changes. All in all, user-tailorable interfaces proved beneficial for takeovers as they could accommodate inter-driver and intra-driver variability. Consequently, they should play a crucial role in future takeover interface design.
