Speakers

Abstract

Tracking anatomy remains a fundamental challenge in soft-tissue surgery, where imperfect, limited sensing, especially for subsurface structures, complicates procedures. We propose a probabilistic, multi-modal fusion framework for real-time anatomical tracking in robot-assisted surgery, combining volumetric tissue simulation with endoscope video, ultrasound, and robot proprioception into a physically grounded scene model.

We initialize a position-based tissue simulation from preoperative CT, then register it to the physical scene using SLAM reconstruction from an initial endoscopic exploration. The registered model predicts tissue deformation from instrument motion; a factor graph then corrects these predictions in real time by fusing them with endoscope, ultrasound, and robot proprioception measurements. The result is continuous, real-time volumetric tracking of tissue deformation throughout surgery, which can guide human surgeons, enable surgical automation, and facilitate human-robot collaboration.

Abstract

Recent advances in foundation models are creating new opportunities for intelligent surgical robots to move beyond task-specific perception toward more generalizable surgical environment understanding and autonomous decision-making. In this talk, I will introduce our recent research on integrating vision-language models, multimodal foundation models, and embodied intelligence into surgical robotics. I will discuss how foundation models can support semantic scene understanding, surgical workflow reasoning, spatial and temporal interpretation, and the generation of actionable representations for robotic autonomy. Particular attention will be given to bridging high-level reasoning with low-level robotic perception and control, as well as addressing uncertainty, safety, and generalization in complex surgical environments. I will also discuss emerging directions toward foundation model-enabled surgical copilots and increasingly autonomous robotic systems.

Abstract

With global surgical volumes estimated by the Lancet Commission to exceed 300 million procedures annually, the need for advances in procedural safety and efficacy is more critical than ever. While minimally invasive techniques, facilitated by high-resolution laparoscopy, have demonstrably improved patient recovery profiles and intraoperative visualization, the inherent complexity of human anatomy demands a new paradigm. Artificial Intelligence (AI) is now fundamentally benefiting surgical care, with the emergence of the "AI copilot" representing a pivotal innovation. In this talk, I will present our latest interdisciplinary research at AI for various perioperative tasks, to support surgical education, intra-operative navigation and even semi-autonomous surgery, via improved surgical scene understanding, reasoning and anticipation. The proposed methods cover a wide range of AI topics about novel learning strategies, such as large vision foundation model, multimodal learning, agentic AI, and robot learning, etc. The challenges, up-to-date progresses and promising future directions of intelligent surgery will also be discussed.

Abstract

Autonomous surgery requires a robot to perform many different tissue interactions, including retraction, dissection, cutting, and suturing, each with its own contact dynamics and failure modes. Most approaches today have focused on large data-driven models that embed task-specific policies with kinematic and mechanical nuances into a latent space, which may be easy to train with more data, but lose explainability or transferability. This talk considers a different approach: where visual flow can be used as a representation for training, learning, and generalizing surgical autonomy across platforms and tasks. I will discuss our approaches that try to achieve this result across a variety of tasks and platforms.

Abstract