HMR-1: Hierarchical Massage Robot with Vision-Language-Model for Embodied Healthcare

A research team from multiple institutions has published a paper on HMR-1 (Hierarchical Massage Robot-1), a novel framework that combines robotics with multimodal AI for embodied healthcare. The system addresses two major gaps in the field: the lack of standardized benchmarks and the scarcity of open-source datasets for acupoint massage. To solve this, the researchers created MedMassage-12K—a comprehensive multimodal dataset containing 12,190 images with 174,177 question-answer pairs covering diverse lighting conditions and backgrounds. This dataset serves as a foundational resource for training and evaluating AI models in therapeutic contexts.

The HMR-1 framework operates through a two-tiered architecture. The high-level module uses multimodal large language models (MLLMs) to interpret natural language instructions and visually identify acupoint locations on the human body. The low-level control module then translates these identified points into precise robotic massage trajectories. The team evaluated several existing MLLMs on their new benchmark and demonstrated the framework's practical viability by fine-tuning Alibaba's Qwen-VL model. Physical experiments confirmed the system's applicability, showing how AI can bridge the gap between language understanding and physical action in rehabilitation settings.