Investigating the Validity of Augmented Reality Simulations for Human-Robot Interaction through Trust

Abstract

Human-Robot Interaction (HRI) research often requires resource-intensive, safety-critical physical setups. Augmented Reality (AR) simulations could offer a viable alternative, but their validity remains unclear. To investigate this, we conducted a between-subjects study comparing trust ratings in an AR-based and a real-world handover task using a physical robotic arm. Participants (N = 40) carried out a series of handovers with the robot and experienced two types of trust violations: the robot dropping the handover-object and handing it over in an inconvenient position. Our results, analysed using Bayesian methods, showed no clear evidence for differences between the AR and physical conditions and, in some cases, even practical equivalence (in correctly executed handovers and during violations where the robot dropped the object). These findings provide initial evidence that AR-based simulations can yield trust-related results comparable to those observed in physical human–robot interactions.