Designing Human-Robot Interaction for service applications

Abstract

As service robots are intended to serve at close range and cater to the needs of human users, Human-Robot Interaction (HRI) has been identified as one of the most difficult and critical challenges in research for the success of service robotics. In particular, HRI requires highly complex software integration to enable a robot to communicate in a manner that is natural and intuitive to the human user. An initial service robot prototype was developed by integrating several existing research projects at the University of Auckland and deployed in a user study. The result showed the need for more HRI abilities to interactively engage the user and perform task-specific interactions. To solve these requirements and deal with relevant issues in software integration, I proposed a design methodology which guides HRI designers from design to implementation. In the methodology, Unified Modelling Language (UML) and an extension, UMLi, were used for modelling a robot's interactive behaviour and communicating the interaction designs within a multidisciplinary group. Notably, new design patterns for HRI were proposed to facilitate communication of necessary cues that a robot needs to perceive, or express during an interaction. The methodology also emphasises an iterative process to discover and design around limitations of existing software technologies. In addition, a component-based development approach was adapted to further help HRI designers in handling the complexity in software integration by modularising the robot's functionalities. As a case study, I applied this methodology to implement a second prototype, Charlie. In a user study with older people (65+) in an aged care facility in New Zealand, the robot was able to detect and recognise a human user in 59 percent of the interactions that happened. Over the two-week period of the study, the robot performed robustly six hours daily and provided assistance in measurement of a range of vital signs. The interaction patterns proposed, were also validated for future reuse. The results indicate the validity of the methodology in developing robust and interactive service applications in real world environments.