Human-robot collaborative teams require measurable indicators of trust to ensure safe, effective, and acceptable operation. Empirical and theoretical work emphasizes multidimensional measurement because trust is not directly observable: it manifests in what people say, how they act, how well teams perform, and how bodies respond. Foundational theory from Raja Parasuraman George Mason University, Thomas B. Sheridan Massachusetts Institute of Technology, and Christopher D. Wickens University of Illinois Urbana-Champaign frames trust as tightly linked to levels of automation and operator reliance, which grounds why multiple metrics are necessary.
Subjective and behavioral metrics
Self-report instruments capture perceived trust and are straightforward to administer, but they reflect beliefs rather than moment-to-moment reliance. Standard questionnaires derived from human–automation research remain useful for comparative studies and for diagnosing attitudes after trials. Behavioral metrics quantify actual reliance: frequency of human interventions, acceptance of robot suggestions, delegation rates, and response latency. Because Parasuraman, Sheridan, and Wickens highlight the role of automation level in shaping human action, behavioral measures serve as direct indicators of operational trust and reveal calibration between human expectations and robot behavior.
Performance, physiological, and model-based measures
Objective performance outcomes such as task completion time, error rates, and recovery from failure indicate the practical consequences of trust or mistrust. Physiological signals—heart rate variability, electrodermal activity, and eye-tracking—offer continuous, implicit measures that can detect stress or cognitive load associated with trust dynamics, though interpretation requires careful context control. Computational trust models, including Bayesian update and probabilistic observer models, convert observed behavior and outcomes into interpretable trust estimates that support adaptive autonomy and can predict future reliance.
Cultural, environmental, and territorial factors modulate what metrics mean in practice. Cultural norms about deference to machines, workplace safety practices, and regulatory environments change baseline willingness to rely on robots; field deployments in extreme environments amplify the consequences of miscalibrated trust. Causes of trust shifts commonly include system reliability, transparency, predictability, and social cues embedded in robot behavior; consequences range from efficient teamwork to dangerous overreliance or disuse.
For robust evaluation, combine subjective, behavioral, performance, physiological, and model-based metrics to triangulate trust across time scales and contexts. This multimodal approach aligns with longstanding human-automation research and provides the evidence practitioners need to design systems that maintain appropriate reliance and protect people and environments.