Understanding and modelling how humans develop and maintain trust in robots is crucial for ensuring appropriate trust calibration during Human-Robot Interaction (HRI). This paper presents a mathematical model that simulates a three-layered framework of trust, encompassing dispositional, situational and learned trust. This framework aims to estimate human trust in robots during real-time interactions. Our trust model was tested and validated in an experimental setting where participants engaged in a collaborative trust game with a robot over four interactive sessions. Results from mixed-model analysis revealed that both the Trust Perception Score (TPS) and interaction session significantly predicted the Trust Modeled Score (TMS), explaining a substantial portion of the variance in TMS. Statistical analysis demonstrated significant differences in trust across sessions, with mean trust scores showing a clear increase from the first to the final session. Additionally, we observed strong correlations between situational and learned trust layers, demonstrating the model’s ability to capture dynamic trust evolution. These findings underscore the potential of this model in developing adaptive robotic behaviours that can respond to changes in human trust levels, ultimately advancing the design of robotic systems capable of real-time trust calibration.

This study explores optimising human-robot trust using reinforcement learning (RL) in simulated environments. Establishing trust in human-robot interaction (HRI) is crucial for effective collaboration, but misaligned trust levels can re-strict successful task completion. Current RL approaches mainly prioritise performance metrics without directly addressing trust management. To bridge this gap, we integrated a validated mathematical trust model into an RL framework and conducted experiments in two simulated environments: Frozen Lake and Battleship. The results showed that the RL model facilitated trust by dynamically adjusting it based on task outcomes, enhancing task performance and reducing the risks of insufficient or extreme trust. Our findings highlight the potential of RL to enhance human-robot collaboration (HRC) and trust calibration in different experimental HRI settings.

Existing research has shown that vocal and non-vocal human cues correlate with human trust and distrust behaviours, suggesting their potential to measure human trust in robots in real-time. However, there is a lack of research in Human-Robot Interaction that integrates vocal and non-vocal cues into a comprehensive model to measure trust. This paper aims to estimate human trust in robots by examining vocal and non-vocal cues differences between trust and distrust states across multiple sessions of collaborative game-based HRI with 40 participants. Our analysis revealed that vocal and non-vocal human cues can indeed predict trust in HRI, with certain facial expressions, facial movements, and pitch being significant factors. Random Forest classifier achieved the highest accuracy (84 %) in classifying trust states, with key features such as facial expressions (fear, angry), facial blendshapes (cheekSquintRight, jawRight), and vocal characteristics (Duration, Harmonicity std) being the most predictive of trust. These findings demonstrate the importance of combining vocal and non-vocal cues for accurate trust measurement and highlight the potential for real-time trust assessment in robotic systems.

Existing work on detecting deception has mainly relied on collecting datasets evolving from contrived user interactions. We argue that naturally occurring deception behaviours can inform more reliable datasets and improve detection rates. Therefore, in this paper, we discuss the findings of two experiments which enabled participants to freely and naturally engage in deceptive and truthful behaviours in a game environment. We collected physiological and oculomotor behaviour (PB, & OB) data including electrodermal activity, blood volume pulse, heart rate, skin temperature, blinking rate, and blinking duration during the deceptive and truthful states. We investigate the changes in both PB and OB across repeated interactions and explore the potential of incremental transfer learning in detecting deception. We found significant differences in electrodermal activity, and skin temperature between deception and non-deception groups in both studies. The incremental transfer learning method with a logistic regression classifier detected deception with 80% accuracy, outperforming previous research. These results highlight the importance of collecting data from multiple sources and promote the use of incremental transfer learning to accurately detect deception in real time.

Existing work has shown that physiological behaviours (PBs) can effectively measure trust. However, there is a limited exploration of using multiple PBs concurrently to calibrate human trust in robots during real-time HRI. Additionally, most datasets are based on one-off interactions or a single context. This project addresses this gap by examining differences in PBs between trust and distrust states and investigating how these PBs change over repeated interactions in different contexts. We conducted two experiments to collect data on electrodermal activity (EDA), blood volume pulse (BVP), heart rate (HR), skin temperature (SKT), blinking rate (BR), and blinking duration (BD) from participants across multiple HRI sessions. The results showed significant differences in HR and SKT between trust and distrust states in Study 1, and significant differences in HR in Study 2. Furthermore, the Decision Tree classifier achieved the highest accuracy of 79% in classifying trust when using the incremental transfer learning algorithm for collective datasets. These results highlight the potential of using PBs for real-time trust measurement in HRI and suggest further exploration of incremental transfer learning methods to enhance trust prediction across different interaction contexts.

Modelling humans’ trust in robots is critical during human-robot interaction (HRI) to avoid under- or over-reliance on robots. Currently, it is challenging to calibrate trust in real-time. Consequently, we see limited work on calibrating humans’ trust in robots in HRI. In this paper we describe a mathematical model that attempts to emulate the three-layered (initial, situational, learned) framework of trust capable of potentially estimating humans’ trust in robots in real-time. We evaluated the trust model in an experimental setup that involved participants playing a trust game on four occasions. We validate the model based on linear regression analysis that showed that the trust perception score (TPS) and interaction session predicted the trust modelled score (TMS) computed by applying the trust model. We also show that TPS and TMS did not change significantly from the second to the fourth session. However, TPS and TMS captured in the last session increased significantly from the first session. The described work is an initial effort to model three layers of humans’ trust in robot in a repeated HRI setup and requires further testing and extension to improve its robustness across settings.

Trust is one of the necessary factors for building a successful human-robot interaction (HRI). This paper investigated how human trust in robots differs across HRI scenarios in two cultures. We conducted two studies in two countries: Saudi Arabia (study 1) and the United Kingdom (study 2). Each study presented three HRI scenarios: a dog robot guiding people with sight impairments, a teleoperated robot in healthcare, and a manufacturing robot. Study 1 shows that participants’ trust perception score (TPS) was significantly different across the three scenarios. However, Study 2 results show a slightly significant variation in TPS across the scenarios. We also found that the relevance of trust for a given task is an indicator of a participant’s trust. Furthermore, the findings showed that trust scores or factors affecting users’ trust vary across cultures. The findings identified novel factors that might affect human trust, such as controllability, usability and risk. The findings direct the HRI community to consider a dynamic and evolving design for modelling human-robot trust because factors affecting humans’ trust are evolving and will vary across different settings and cultures.