TY - GEN
T1 - Autonomy for Surgical Robot Systems
AU - Schwaner, Kim Lindberg
PY - 2021/12/13
Y1 - 2021/12/13
N2 - Robotic systems play an increasingly important role in minimally invasive surgical interventions. Robot-assisted minimally invasive surgery is performed by surgeons remotely controlling instruments from a console. In this context, a "surgical robot" is not a robot in the sense that it can carry out actions autonomously. However, introducing autonomous capabilities into robot-assisted surgery has potential benefits, including high and consistent quality of procedures and reducing the surgeon’s workload and fatigue. This may ultimately increase hospital throughput to the benefit of both the patient and society.This thesis addresses three technological challenges to automating surgical interventions. One pertains to the inaccuracies of surgical robot systems, a second to the difficulty of planning tasks in dynamically changing soft-tissue environments and a third to the limited sensing capabilities of surgical robots.To address the first challenge, we first implemented a cascade controller for the Raven-II surgical robot. Despite improved trajectory-following, the accuracy of the Raven-II was still not satisfactory for automating surgical tasks. As an alternative, we developed an entirely new platform for non-clinical surgical robotics research. Through a series of small experiments, the platform was demonstrated to be feasible for surgical task automation.Regarding planning of surgical tasks, the problem was approached with the idea that the robot should learn from the surgeon. With this in mind, we applied a learning from demonstration scheme to automate surgical suturing – a task which frequently occurs in surgery. To encode learned motions and adapt them to new situations, we turned to Dynamic Movement Primitives (DMPs). This was also used in a human- robot collaborative task, where the motion of one manipulator was automated, based on the motion of manually controlled manipulator.Finally, to improve sensing capabilities of surgical robots, this thesis explores two avenues. One is using existing visual feedback and applying a convolutional neural network to locate a suture needle and surgical instruments. A second is integrating bio-electrical sensors into existing surgical instruments. This can potentially assist the surgeon with locating pathological tissue or function as an additional sensor input to an autonomous system.
AB - Robotic systems play an increasingly important role in minimally invasive surgical interventions. Robot-assisted minimally invasive surgery is performed by surgeons remotely controlling instruments from a console. In this context, a "surgical robot" is not a robot in the sense that it can carry out actions autonomously. However, introducing autonomous capabilities into robot-assisted surgery has potential benefits, including high and consistent quality of procedures and reducing the surgeon’s workload and fatigue. This may ultimately increase hospital throughput to the benefit of both the patient and society.This thesis addresses three technological challenges to automating surgical interventions. One pertains to the inaccuracies of surgical robot systems, a second to the difficulty of planning tasks in dynamically changing soft-tissue environments and a third to the limited sensing capabilities of surgical robots.To address the first challenge, we first implemented a cascade controller for the Raven-II surgical robot. Despite improved trajectory-following, the accuracy of the Raven-II was still not satisfactory for automating surgical tasks. As an alternative, we developed an entirely new platform for non-clinical surgical robotics research. Through a series of small experiments, the platform was demonstrated to be feasible for surgical task automation.Regarding planning of surgical tasks, the problem was approached with the idea that the robot should learn from the surgeon. With this in mind, we applied a learning from demonstration scheme to automate surgical suturing – a task which frequently occurs in surgery. To encode learned motions and adapt them to new situations, we turned to Dynamic Movement Primitives (DMPs). This was also used in a human- robot collaborative task, where the motion of one manipulator was automated, based on the motion of manually controlled manipulator.Finally, to improve sensing capabilities of surgical robots, this thesis explores two avenues. One is using existing visual feedback and applying a convolutional neural network to locate a suture needle and surgical instruments. A second is integrating bio-electrical sensors into existing surgical instruments. This can potentially assist the surgeon with locating pathological tissue or function as an additional sensor input to an autonomous system.
U2 - 10.21996/x9jj-8z21
DO - 10.21996/x9jj-8z21
M3 - Ph.D. thesis
PB - Syddansk Universitet. Det Tekniske Fakultet
ER -