Social robots will become increasingly widespread in the coming decades. So far, they have only been introduced in a few real-world environments, including some airports, museums, and hospitals. Dr Xavier Alameda-Pineda at Inria, the French National Institute for Research in Digital Science and Technology, and all the partners of the Horizon 2020 SPRING project recently set out to evaluate the potential of social robots in the context of geriatric healthcare. Read More
They carried out their research at the Broca gerontology hospital, which specialises in diagnosing and treating older adults with signs of dementia. Social robots could potentially interact with these patients as they wait for their appointments. The research team implemented a new software that supports interactions between robots and patients, and deployed it in a 1.65-metre-tall robot called ARI, developed by PAL Robotics.
ARI is a humanoid body placed on a mobile base, with a touchscreen on its torso, a camera in its head, two fish-eye cameras on its chest and back, four microphones, and eyes with integrated LCD screens that can reproduce human eye movements.
In order to provide it with abilities relevant to its missions, the team produced state-of-the-art algorithms for environment and human behaviour analysis as well as multi-party conversation; all relying on the fusion of audio and visual inputs from ARI’s sensors.
The team deployed the robot at the gerontology hospital during two different experimental periods – one running from May to July 2023, and the other from September 2023 to January 2024. Before the second trial, the robot’s software was modified to include large language models and more robust speech recognition models, which enable sophisticated dialogue between ARI and humans.
In total, 63 patients interacted with the robot. The participants then completed a questionnaire that assessed the extent to which they accepted the robot and its user-friendliness.
The usability ratings that patients gave the robot were similar between the first and second experimental trial. Interestingly, participants in the second trial accepted the robot more than those in the first trial. This difference could be due to the addition of large language models before the second trial, which improved the robot’s responses.
In their feedback, 40% of participants said that the robot would be best suited for entertainment, 30% for promoting social interactions without health-related risks, 23% for helping visitors to prepare for medical consultations, 5% for reception and welcoming, and 2% for orientation and guidance.
The team’s study suggests that social robots could be valuable additions to geriatric healthcare facilities, while also highlighting the benefits of large language models to enable smoother interactions between humans and robots. The team’s results also pinpoint some particularly promising uses for these robots, suggesting that they could be valuable for entertaining guests and preparing them for medical appointments.
