**Background:** With increasing life expectancy, societies face challenges related to ageing populations and the need for smart living solutions to support older adults. The e-VITA project is a three-year joint European (H2020) and Japanese (MIC) research project (grant no. 101016453 and JPJ000595) developing a virtual coach to help older persons manage their own health and daily activities. The coach provides support in cognition, physical activity, mobility, mood, social interaction, leisure, and spirituality using individualized profiling, personalized recommendations, big data analytics, social-emotional computing, non-intrusive sensors, and natural interactions via 3D-holograms, emotional objects, or robotic technologies.
**Methods:** Requirements were gathered through semi-structured interviews with 58 older community-dwelling adults (aged 65+) in Germany (n=11, mean age 70, range 65–79), Italy (n=5, mean age 70), France (n=12, mean age 74.2), and Japan (n=30, mean age 71.1, std 2.96; 19 female, 11 male). Participatory design studies were conducted in Living Labs in Germany (13 participants, ages 65–86) and Japan (4 participants, ages 70–76). The prototype system integrated: (1) multimodal data fusion from sensors (worn, environmental, smart home) using an architectural approach based on FIWARE standards, with machine learning algorithms (CNN-LSTM) for actimetric analysis; (2) an Emotion Detection System (EDS) using a 2D-CNN + LSTM model on log-mel spectrograms achieving average F1 scores of 62% (German), 59% (Italian), and 47% (Japanese); (3) knowledge sources including knowledge graphs (Neo4j) and JSON knowledge bases; and (4) a Rasa Open Source Conversational AI dialogue system using DIET for NLU and TED for response generation. The feasibility study tested the prototype in real home conditions with 15 users in Europe (mean age 71.9) and 5 users in Japan (aged 67–91). Eligibility was assessed using MOCA, GDS, and SPPB. Adherence was recorded automatically; user experience was measured via questionnaires (5-point Likert scale) and semi-structured interviews.
**Key Results:** In Europe, interactions with the Gatebox device far exceeded those with the Nao robot. Italy had the highest number of interactions among European countries. In Japan, user responses to questionnaires were generally positive, indicating that dialogue contents were applicable to daily living and interaction was easy. Users found actual interactions better than expected and felt the robot was a good communicator. However, overall results from European feasibility studies revealed rather negative opinions towards the technology, though participants did not show embarrassment or nervousness. Many users found questions related to user profiling and preferences uncomfortable. Cultural background was emphasized as important for dialogue design. The NLU accuracy for English improved by 5.81% over vanilla Rasa after adding a grammar/spell checker and better dense featurizer. Users recommended improvements including longer dialogues, better handling of unrecognized inputs, and new features (entertainment, reminders, seasonal information, emergency contacts).
**Clinical Implications:** The e-VITA virtual coach represents a promising but early-stage approach to promoting active and healthy ageing through technology. The feasibility study highlighted significant usability and acceptability challenges that must be addressed before clinical deployment. The negative opinions in Europe suggest that current technology may not yet meet user expectations or integrate seamlessly into daily life. The planned randomized controlled Proof of Concept study with 240 participants across 4 countries over 6 months will provide more robust evidence on effectiveness. Key areas for improvement include dialogue accuracy, cultural tailoring, longer conversational capabilities, and integration of sensor data for personalized coaching. The project's focus on multimodal sensing, emotion detection, and knowledge graphs offers a comprehensive framework, but real-world adoption will depend on addressing privacy concerns, user comfort with profiling, and demonstrating tangible benefits for independent living and social inclusion.