Luke Liu, Moses Garuba, Ph.D., LL.M

DOI: 10.5281/zenodo.20517610

Social robots are increasingly deployed in elderly care to provide companionship, emotional support, and assistive monitoring. AI-based vision models have enabled these robots to detect and interpret facial features in support of personalized engagement, but many current systems rely on cloud-based architectures that raise privacy and ethical concerns around the transmission of sensitive biometric data. Beyond interception risk, algorithmic bias stemming from insufficient dataset diversity may produce inferences that perform unequally across demographic groups. This study proposes Edge AI vision as a privacy-preserving framework for elderly care robotics. An Edge AI vision model was developed using Edge Impulse and trained to identify the eyes and mouth of male and female elderly subjects aged 55–94. The model achieved classification accuracy of 98.04% on male subjects and 99.51% on female subjects, with no statistically significant difference between groups (z = 1.43, p = 0.15), suggesting that edge-based inference can deliver equitable and privacy-preserving vision capabilities for elderly care robotics.

edge artificial intelligence, social robotics, elderly care, AI vision, Edge Impulse, FOMO, MobileNetV2, facial feature detection, algorithmic fairness, privacy.

This study proposed and evaluated an Edge AI vision model for elderly social assistance robots, developed using Edge Impulse and trained on a curated subset of AGFW (ages 55–94). The model, built on the FOMO MobileNetV2 0.35 architecture, achieved 98.04% accuracy on male subjects and 99.51% on female subjects, with the difference failing to reach statistical significance (z = 1.43, p = 0.15) at α = 0.05. Two conclusions follow: first, the technical feasibility of edge-deployable vision for elderly care robotics, demonstrated at the scale of feature localization on a constrained model, supports a broader architectural shift away from cloud-based vision pipelines and toward on-device inference; second, the absence of a detectable male/female accuracy disparity is encouraging evidence that lightweight, edge-deployed vision systems can be both privacy-preserving and demographically equitable, though equitability along other axes remains to be characterized. Continued work along the directions outlined in the previous section will be necessary to translate this feasibility result into a deployed system that can responsibly support elderly users in their daily lives.

M. Nielsen, "The evolution of assistive technology for seniors," DEV Community, Jan. 2025. [Online]. Available: https://dev.to/vela-chairs/the-evolution-of-assistive-technology-for-seniors-58eb
N. L. A. Rashid, Y. Leow, P. Klainin-Yobas, S. Itoh, and V. X. Wu, "The effectiveness of a therapeutic robot, ‘PARO’, on behavioural and psychological symptoms, medication use, total sleep time and sociability in older adults with dementia: A systematic review and meta-analysis," Int. J. Nurs. Stud., vol. 145, p. 104530, Sept. 2023.
T. Huang and C. Huang, "Attitudes of the elderly living independently towards the use of robots to assist with activities of daily living," Work, vol. 69, pp. 55–65, May 2021.
O. O. Aramide, "Edge AI and its impact on resilient AI fabric design: Distributed intelligence and data locality," SAMRIDDHI J. Phys. Sci., Eng. Technol., vol. 17, pp. 12–24, 2025.
Andromeda Robotics, "Andromeda | Personalised robot companions for aged care home residents in Australia," 2025. [Online]. Available: https://andromedarobotics.ai/
J. Liu, X. Wang, and J. Zhang, "Investigating elderly individuals’ acceptance of artificial intelligence (AI)-powered companion robots: The influence of individual characteristics," Behav. Sci., vol. 15, p. 697, May 2025.
B. Deusdad, "Ethical implications in using robots among older adults living with dementia," Front. Psychiatry, vol. 15, Sept. 2024.
L. Battistuzzi, A. Sgorbissa, C. Papadopoulos, I. Papadopoulos, and C. Koulouglioti, "Embedding ethics in the design of culturally competent socially assistive robots," IEEE, Feb. 2020.
V. Karami, M. J. Yaffe, G. Gore, A. Moon, and S. Abbasgholizadeh Rahimi, "Socially assistive robots for individuals with Alzheimer’s disease: A scoping review," Arch. Gerontol. Geriatr., vol. 123, pp. 105409, Mar. 2024.
M. Leineweber, C. V. Keusgen, M. Bubeck, J. Haltaufderheide, R. Ranisch, and C. Klingler, "Ethical aspects of the use of social robots in elderly care – a systematic qualitative review," arXiv:2505.09224, 2025.
C. Moro, S. Lin, G. Nejat, and A. Mihailidis, "Social robots and seniors: A comparative study on the influence of dynamic social features on human robot interaction," Int. J. Soc. Robot., vol. 11, pp. 5–24, Aug. 2018.
J. A. Dosso, J. N. Kailley, G. K. Guerra, and J. M. Robillard, "Older adult perspectives on emotion and stigma in social robots," Front. Psychiatry, vol. 13, Jan. 2023.
K. Alikhademi, E. Drobina, and J. D. Louis, "Person detection through the lens of algorithmic bias," OpenReview, 2025. [Online]. Available: https://openreview.net/forum?id=tC1b9DBWww
C. Wang, Z. Yuan, P. Zhou, Z. Xu, R. Li, and D. O. Wu, "The security and privacy of mobile-edge computing: An artificial intelligence perspective," IEEE Internet Things J., vol. 10, pp. 22008–22032, Dec. 2023.

Luke Liu, Moses Garuba, Ph.D., LL.M (2026). An Edge AI-Vision Model for Elderly Social Assistance Robots Using Edge Impulse. International Journal of Computer Techniques, 13(3), 822–831. ISSN: 2394-2231. DOI: 10.5281/zenodo.20517610