Recently developed AI-driven human-robot interaction (HRI) systems using RNNs, LSTM networks, and Transformer-based models like BERT and GPT, have shown the potential to enhance geriatric care. These technologies use sensor and IoT data to provide individualized, responsive, and effective assistance for elderly people. Many obstacles prevent these systems from reaching their full potential. Training data quality and representativeness are major issues. Collecting broad and reliable datasets that represent older people's real-world experiences is vital yet challenging. The ethical and technological challenges of protecting senior users' sensitive data are very high. To offer accurate and contextually relevant replies, multimodal data from aural, visual, and tactile sources must be seamlessly integrated. Maintaining robots' physical and cognitive ergonomics for natural and intuitive interactions is another design challenge. This study offers an AI-driven HRI system to improve senior care to solve these issues. Creating powerful AI models that can learn from high-quality, diversified datasets and combining multimodal data processing approaches to increase robotic caregiver accuracy and responsiveness are the goals. To employ the robots safely and ethically in geriatric care, the project will improve their physical design and apply strong data privacy and security safeguards. The suggested approach addresses these important difficulties to improve older well-being and independence by providing more individualized care and assistance. © 2024 Elsevier B.V., All rights reserved.