Decision support for partial automation: Driver behavior and trust with different versions of automated lane change

Automated lane change systems allow drivers to receive support for maneuver execution and sometimes initiation. Research suggests that delegating action selection to automation can impair awareness and reduce readiness to intervene if automation fails. The current study was designed to compare manual lane changes with two automated variants: driver-initiated but system-executed, and system-initiated and -executed after driver confirmation.

Fourteen drivers were provided with instrumented Tesla Model 3s for a 4-week study period, during which video of the driver, the cabin, and the forward roadway, as well as data on vehicle speed and position were recorded. Automated lane changes were compared with manual lane changes, with each automated event matched to a similar manual one through stratified random sampling. Self-report questionnaires were administered at the conclusion of the study to examine drivers' subjective experiences.

Drivers undergoing both variants of automated lane changes appeared to use the information displayed on the vehicle's center stack as a replacement for directly observing the forward roadway. The system-initiated automated lane changes were unique, however, in that drivers exhibited even shorter glances to the side mirrors and more frequent cell phone use compared with the other control modes, as well as reduced on-road glance duration over the second half of the study period. We also found that driver mistrust increased the most when the vehicle canceled an automated lane change, with system cancellations predicting lower future system use and less positive subjective perceptions of the system overall.

Findings clarify differences between automation that supports the execution of a maneuver compared with one that supports both initiation and execution. The latter may risk driver disengagement and insufficient automation verification behavior.

Automakers and system designers should be cautious about implementing automated support for action selection without ensuring that drivers engage in sufficient oversight.