We explore the potential of artificial intelligence-based robotic systems to address the increasing demand for caregiving services driven by global demographic changes, aging populations, war actions, natural disasters, and pandemics. We apply a system approach to problem-solving, where analysis and scenario planning enable the creation of an effective model for implementing innovations in the medical field. We highlight the limitations of traditional robotic systems, particularly their dependence on costly hardware and sensors, and propose leveraging advancements in deep learning to develop safe and cost-effective caregiving robots. Specifically, we focus on the application of neural-network-based policies trained with reinforcement and imitation learning to enable adaptable, real-time human-robot interactions. Our approach aims to reduce hardware costs while maintaining high safety standards, addressing tasks such as feeding, bathing, and rehabilitation. We also focus on safety monitoring systems based on deep learning, which provide early anomaly detection capabilities based on patient’s and robot’s behavior. A composite control system should include both neural net policies and the safety models, making it suitable for applications with human-robot interaction component. By advancing methodologies for validating and deploying these systems, we contribute to the emerging field of assistive robotics and robot-assisted rehabilitation, with the potential for transformative impacts in caregiving, directly impacting the national security of countries employing such technology, positively affecting healthcare economics, quality, availability, and bringing sociological benefits.