A Virtual Assistive Companion for Older Adults: Design Implications for a Real-World Application

Socially intelligent virtual agents are a promising solution to the increasing challenges of eldercare. This article describes an autonomous conversational agent system, simulating human-like affective behaviour to act as a daily life companion for older adults living at home. We present results from a user-centred design study that informed the design of the companion. The interaction with the user is based on a multimodal interface including automatic speech recognition, text-to-speech, and a graphical touch-based user interface. The companion offers support to older adults by locating objects, offering reminders, and guidance with household activities. It also supports emergency detection and handling. We present the companion's architecture and key components as well as the affective interaction paradigm and its implementation in the multimodal user interface. We conducted an exploratory evaluation study where the companion was introduced in 20 homes of older adults (aged 65+) in two EU countries. For a total duration of 12 weeks, various scenarios were tested, covering different types of interactions and tasks that occur in daily life. Qualitative and quantitative data on acceptance, perceived usability, and usefulness yield rich information on how agents perform in real-world settings and which factors influence their success in this context. We reflect on the results of the evaluation study in terms of lessons learned and discuss future opportunities for fellow researchers who are striving to bring virtual agents out of the laboratories into successful real world applications.