Research Topics
Our research sits at the interdisciplinary nexus of machine learning, robotic surgical intelligence and medical image analysis, aiming to develop innovative intelligent systems that advance diagnosis, intervention, and medical education through next-generation healthcare technologies. Our work spans embodied AI for surgical robotics, medical image understanding, robot sensing and planning in dynamic environments, agentic AI systems for clinical decision-making, LLM for education and smart XR for medical training. Representative contributions include 3D deep learning for high-dimensional image computing, domain adaptation and generalization across heterogeneous medical data, surgical video analysis with efficient spatio-temporal learning, and visual-kinematic perception and automation in surgical robotics.
Recent focus: 1) Surgical Robotics, 2) Agentic AI Systems for Healthcare Applications, 3) Human Robots for Elderly Care, 4) Safe Embodied AI, 5) LLMs and Smart XR for Medical Education
Surgical Robotics
Agentic AI Systems for Healthcare Applications
Humanoid Robots for Elderly Care
Safe Embodied AI
LLMs and Smart XR for Medical Education
The future of robotic surgery lies in the integration of artificial intelligence and robotic systems that can augment surgeons' capabilities while improving patient outcomes. Our research in surgical robotics addresses the critical need for intelligent, autonomous systems that can transform medical education, enhance surgical efficiency and safety through advanced machine learning techniques and embodied intelligence.
Our research spans from intelligent data analysis to autonomous task execution in robotic surgery. We will develop multi-modal analysis systems to understand activities and scenes at various granularities, enabling semantic and depth-aware perception in robotic surgery. Domain-specific large vision-language models and augmented reality systems will be created to advance surgical education and provide real-time intraoperative guidance to surgeons. Our work will further include building realistic surgical robot simulators and developing sophisticated policy learning approaches, including imitation and reinforcement learning algorithms. These integrated components will work together to establish a robust autonomous framework capable of performing surgical tasks for human-robot collaboration. Ultimately, these comprehensive efforts will create a new paradigm of robotic surgery that can reduce surgeons' burden while improving patient outcomes.
Agentic AI Systems for Healthcare Applications: This research topic aims to develop autonomous, multi-agent AI systems that integrate clinical and patient data to enable end-to-end automation of healthcare workflows. We aim to pioneer multimodal, multi-agent AI systems to revolutionize clinical intelligence by integrating diverse data streams—including medical imaging, genomic profiles, electronic health records, and biomedical literature—into collaborative AI networks.
These systems deploy specialized agents (e.g., diagnostic, predictive, and evidence-synthesis agents) that dynamically interact to enhance precision medicine, accelerate disease detection, and generate patient-specific therapeutic insights. The core aim is to overcome data fragmentation and clinical complexity through coordinated AI cognition, ultimately advancing diagnostic accuracy, treatment personalization, and real-time decision support in high-stakes healthcare scenarios.
Humanoid Robots for Intelligent Assistance in Elderly Care: The accelerating trend of global population aging presents a pressing societal challenge: a rapidly growing demand for effective elderly care solutions that can seamlessly integrate into human-centric environments. We posit that humanoid robots, with their anthropomorphic form, are uniquely suited to this role, as they are designed to navigate spaces and use tools originally intended for people.
Our research agenda is therefore dedicated to creating intelligent humanoid assistants for these scenarios. We will focus on advancing the core capabilities essential for this vision, including robust whole-body control for mobility and physical support in cluttered homes. We will investigate bimanual dexterous manipulation to perform complex, collaborative tasks, and tackle the nuanced challenge of deformable object manipulation for handling items like clothing and bedding. Moreover, these physical competencies are underpinned by intuitive human-robot interaction and collaboration, ensuring the robot can understand and partner with its user. The ultimate objective is to provide intelligent, personalized assistance with daily activities, professional nursing support, and meaningful companionship for elderly individuals, thereby enhancing their quality of life.
Safe Embodied AI for General-Purpose Robot Manipulation: As robotic systems become increasingly prevalent in various scenarios of our daily life, ensuring their safety and reliability has emerged as a crucial research frontier. Our research will explore cutting-edge safe embodied AI technologies for robot manipulation.
Specifically, we will study spatial-aware vision-language models (VLMs) to improve the robot's 3D understanding, enabling reliable task planning that is grounded in a deep awareness of object spatial relationships and the ability to satisfy precise spatial constraints. For physical interaction, we will investigate advanced multi-modal policy learning methods that fuse language instructions with visual and tactile feedback, enabling adaptive grasping with precise and safe force control. Moreover, we will study advanced reinforcement learning methods to fine-tune vision-language-action models (VLAs), enhancing their control robustness in the presence of various environmental disturbances and sensor noises. The ultimate objective is to develop intelligent robot systems with enhanced safety and robustness in planning, grasping, and motion control for general-purpose robot manipulation.
LLMs for Medical Education: We aim to reform and innovate research, applications, and design to advance the next generation of medical education. Our focus lies in creating a new paradigm for human-computer interaction in medical teaching, enabling medical students worldwide to access higher-quality, more reliable, and effective education and resources. Beyond professional medical training, we are equally committed to promoting public health education. By designing new interactive media and information dissemination methods, we strive to deliver health knowledge and information to people worldwide—particularly in underserved and underdeveloped regions—in a more accessible and impactful way.
Smart XR: This research topic focuses on developing intelligent XR (AR/VR) systems to enhance medical education and surgical assistance through immersive simulation and spatial computing. We aim to advance clinical training and intraoperative support by integrating patient-specific data, 3D anatomical modeling, and interactive visualization into adaptive XR environments. These systems combine real-time imaging, procedural guidance, and intuitive user interaction to improve anatomical understanding, skill acquisition, and surgical precision. The goal is to bridge the gap between learning and practice by delivering context-aware, task-specific XR tools that elevate training quality, reduce surgical risk, and support decision-making in complex clinical settings.