Trust in an Autonomously Driven Simulator and Vehicle Performing Maneuvers at a T-Junction with and Without Other Vehicles

Abstract

Autonomous vehicle (AV) technology is developing rapidly. Level 3 automation assumes the user might need to respond to requests to retake control. Levels 4 (high automation) and 5 (full automation) do not require human monitoring of the driving task or systems [1]: the AV handles driving functions and makes decisions based on continuously updated information. A gradual switch in the role of the human within the vehicle from active controller to passive passenger comes with uncertainty in terms of trust, which will likely be a key barrier to acceptability, adoption and continued use [2]. Few studies have investigated trust in AVs and these have tended to use driving simulators with Level 3 automation [3, 4]. The current study used both a driving simulator and autonomous road vehicle. Both were operating at Level 3 autonomy although did not require intervention from the user; much like Level 4 systems. Forty-six participants completed road circuits (UK-based) with both platforms. Trust was measured immediately after different types of turns at a priority T-junction, increasing in complexity: e.g., driving left or right out of a T-junction; turning right into a T-junction; presence of oncoming/crossing vehicles. Trust was high across platforms: higher in the simulator for some events and higher in the road AV for others. Generally, and often irrespective of platform, trust was higher for turns involving an oncoming/crossing vehicle(s) than without traffic, possibly because the turn felt more controlled as the simulator and road AVs always yielded, resulting in a delayed maneuver. We also found multiple positive relationships between trust in automation and technology, and trust ratings for most T-junction turn events across platforms. The assessment of trust was successful and the novel findings are important to those designing, developing and testing AVs with users in mind. Undertaking a trial of this scale is complex and caution should be exercised about over-generalizing the findings.