Red or Blue Door: Exploring the Behavioural Consequences of Trust Violations due to Robot Error or Choice Using a VR Maze

Human-robot collaboration offers significant benefits as long as humans understand and trust their robotic partners appropriately. Traditionally, a robot's trustworthiness was primarily determined by its performance, but as robots gain autonomy and decision-making authority, collaborators should also pay attention to the robot’s priorities and goals. Losses of trust may arise not just from a robot’s failure (i.e., a performance-based trust violation), but may increasingly stem from misaligned priorities between the trustor and the trustee (i.e. a moral-based trust violation). This study aimed to investigate the effects of (1) trust violations due to a robot’s error vs. choice and (2) prior knowledge of the robot's abilities or intentions on the development of trust (i.e., repeated violations and repair) and compliance. How does the nature of a trust violation (error vs. choice) influence the development of trust and to what extent can prior knowledge about the robot’s abilities or intentions mitigate some of the potential negative effects of a violation? Inspired by earlier HRI paradigms, we developed a Virtual Reality (VR) maze that simulated a military search-task where participants (n = 75) collaborated with an autonomous drone. In a series of rooms, participants could choose to follow or disregard the robot's advice. Due to low compliance rates, we could not conduct our planned analyses. Therefore, we only describe the compliance rates and qualitative feedback. While we were yet unable to conclusively address our research question, the aim of this paper is to share our ideas and insights that can help
to improve the research paradigm.