Abstract
In order to assess trust within domains necessitating Human-Robot Interaction (HRI), such as social and assistive robotics, a multifaceted cognitive infrastructure is essential to facilitate shared knowledge among the participants. This knowledge encompasses information pertaining to the behaviour, beliefs, intentions, and situational awareness of the participants, including assessments of vulnerability, reliability, and risk. The developed knowledge takes the form of a domain-agnostic ontology with multiple layers, allowing for reusability across diverse HRI tasks, and resulting in actions that can be undertaken by a participant to establish trust with the other party. This research introduces the concept of Trust-Inferring Infrastructure for Cognitive Knowledge (TICK) for domains requiring HRI, comprising three key constructs: performance, process, and purpose. Subsequently, the proposed infrastructure is evaluated through real-world scenarios involving humans as trustors and robots as trustees. To validate TICK in these real scenarios, a multimodal sensory module has been integrated into the reasoning mechanism, enhancing the robots' capabilities in understanding human intentions, perceiving the current situation, offering advice to humans, personalising their behaviour to build human trust, and evolving the robot's trustworthiness based on its performance during interactions. Trust is automatically assessed based on proximity and adherence to the robots' recommendations. Furthermore, to quantitatively assess the scalability and flexibility of the proposed approach, a series of experiments were conducted involving a kitchen domain handover task between humans and a table-top robotic arm in a lab setting, encompassing various types of objects and different scene contexts.