Understanding automation handoff impacts on workload and trust when mitigated by reliability displays

Abstract

Current commercial vehicles are beginning to include automated features such as adaptive cruise control and automated lane keeping. This is a first step towards full vehicle automation which is predicted to be possible within the next five years. As automated features are integrated into vehicles, the driver must know how to properly interact with and trust these systems. A key element of drivers interacting and relying on these systems is the handover of control between the vehicle and driver. This handover, occurring during times of automation error, will be a critical point of high workload for drivers when driving a partially or fully automated vehicle. If the driver is aware of the system’s performance and can appropriately calibrate his or her trust, then these instances of handover may become less stressful and easier to complete successfully. This study explored the driving performance, trust, visual scanning behaviors, perceived workload, and objective workload for handover scenarios. There were four between-subjects display conditions: (1) no display; and reliability displays using (2) quantitative information (percentage of reliability); (3) qualitative information (direct representation of a number); and (4) representational information (abstract representation of a number). Participants completed two drives. The first drive aided in familiarization with the automated lane keeping system. In the second drive, the handover drive, participants experienced an automation failure resulting in transition of control from automated to manual. Results from this study showed that there was a difference in subjective experience between the baseline and handover drive due to experiencing an automation failure. Participants in the no display condition were more affected by the automation failure, greatly decreasing their overall trust in the automated lane keeping system. Participants with reliability displays were able to appropriately calibrate their trust to system performance and were less impacted by the automation failure, experiencing a slight, statistically insignificant, decrease in trust. These findings will impact the implementation and design of automation reliability displays and shows that drivers with reliability displays are less impacted by automation failure than those without reliability displays.