Cognitive Robotics for Surgery—Elevating Efficiency in the OR

The increasing use of robots has changed surgical practice in recent years. Precisely controllable instruments and integrated sensor technology enable minimally invasive operations and provide medical staff with additional - often processed - information. Following today’s standard-of-care surgical robots with rigid or actuated instruments, such as the Da Vinci or the Senhance systems, controllable and flexible continuumrobots are now being explored. These compliant robotic instruments are designed to navigate with high precision through the body and around obstacles while reducing the risk of tissue damage. Regardless of their kinematic structure, the next generation of surgical robots will learn from human experts and medical data. This paves the way for context-sensitive, cognitive learning robots that can perceive their environment, learn from surgeons, and assist (semi-) autonomously during surgical steps, enabling true and natural co-operation with human surgeons.<br/><br/>To make this vision a reality, we focuses on both structure and cognition of next generation surgical robots. We utilize machine learning methodologies to provide learning software for cognition-guided robotic assistance during surgery. Further, we investigate minimally invasive and sensorized continuumrobots to provide inherently safe and dexterous smart surgical instruments.<br/><br/><b>Cognitive Robotic Assistance</b><br/>Surgical robots are entering the market with varying degrees of automation. However, (so far) no system exists that operates fully autonomously. The next generation of robotic assistants is represented by cognitive surgical robots that understand their environment and provide context-sensitive support to a human surgeon as required. Natural partnership between a human and a robotic assistant gives rise to the idea that every team member (human or artificial) does what they do best. Human abstraction skills and creative solution finding is assisted by robotic precision and fast interpretation of surgical data, adding value to both the surgeon and the patient.<br/>Machine learning techniques allow a robot to learn surgical behavior during laparoscopic surgery, or in catheter navigation. In particular, we explore reinforcement learning and imitation learning methods where multiple decentralized agents share a goal and cooperatively learn a common task through training or from expert demonstrations. This methodology enables a human operator to take over control of one or multiple instruments during critical surgical phases or during situations with high policy uncertainties.<br/><br/><b>Sensorized flexible Instruments</b><br/>As surgical robots with rigid or articulated instruments became standard of care in many hospitals, international research moved into a new direction. Flexible robotic systems and steerable catheters have been explored for use in otherwise hard-to-reach organs and regions of the body. Flexible continuumrobots allow interaction with soft tissue in several clinical applications, as the compliant structure of these robots minimizes the risk of damage.<br/>Due to the often highly complex kinematics of long and flexible continuumrobots, their control is not trivial. Analytical approaches to modeling and controlling these continuous structures with increasingly high numbers of degrees of freedom are currently facing limitations. Therefore, our approach to extend and combine classical methods with advanced path planning and machine learning holds the potential for precisely controllable flexible smart robotic instruments for surgery that are designed to adapt to their environment inside the human body. By collecting motion data from continuumrobots and integrated sensors, this methodology renders it possible to predict the precise motion and counteract, if required. In the long term, surgical continuumrobots and flexible steerable instruments hold the potential for even less invasive surgery and promise efficiency and safety for surgeons and patients.