Robot-Assisted Tele-Surgery for Tele-Health
Tele-surgery over longer distances is not yet possible. Time-delays, although brief, pose a major challenge and can lead to mechanical instabilities for such remote tele-operation that can disrupt procedures and injure patients. A promising alternative is Supervisory Control, where the remote surgeon supervises each step by specifying subtasks and parameters that are then performed autonomously by the robot, thereby avoiding instabilities. Robot-Assisted Tele-Surgery combines the expertise and intuition of the remote surgeon to supervise with the precision of the local robot system. Currently, there are two barriers to research in Robot-Assisted Tele-Surgery: Researchers do not have access to the control interfaces needed to experiment, and programming of autonomous subtasks that involve deformation of soft tissues and nonlinear dynamics is extremely difficult.
CITRIS researchers have been given full access to the interfaces of the Raven Surgical Tele-Operation System which was designed, fabricated, and interfaced by Prof. Jacob Rosen at UCSC with Blake Hannaford at University of Washington. The researchers will apply statistical robot learning to acquire control policies based on example trajectories provided by human experts. A paper by team leaders Professors Pieter Abbeel and Kenneth Goldberg on the preliminary results won the Best Medical Robotics Paper at the prestigious IEEE International Conference on Robotics and Automation. Working with Doug Boyd from UC Davis, a world-class surgeon, they will implement a system and perform a series of experiments to establish significant proof of concept results for supervised tele-surgery.
This research team brings together established leaders in surgery, robot hardware, robot learning, medical robotics, and automation. Each of the team members has published widely in their areas. The CITRIS seed funding will provide a unique opportunity for leveraging each of their strengths to establish the foundation for a revolution in Tele-Surgery.
The Raven is a state-of-the-art open architecture (software & hardware) surgical robotic system. A Raven system including four arms is currently available at the Bionics Lab (UCSC), and an identical system will be had been installed at UC Berkeley in 2012 as part of an NSF equipment grant which is funding fabrication of 8 identical Raven systems and distributing them among leading groups across the US (Harvard, Hopkins, U. Nebraska, UCLA, UCB, UCSC, UW).
Apprenticeship Learning, is a new statistical approach to robot learning that has the potential to allow robotic surgical assistants to autonomously execute specific subtasks with superhuman performance in terms of speed and smoothness. In the first step, the researchers record a set of trajectories using human-guided back-driven motions of the robot. Their algorithm then analyzed them to extract a smooth reference trajectory, which they execute at gradually increasing speeds using a variant of iterative learning control. They have evaluated this approach on two representative tasks using the Berkeley Surgical Robots: a figure eight trajectory and a two handed knot-tie, a tedious suturing sub-task required in many surgical procedures. Although the Berkeley Surgical Robots are over a decade old and often malfunction, they were able to obtain preliminary results for simple trajectories trained by non-experts. Results suggest that the approach enables (i) rapid learning of trajectories, and (ii) motion that is moderately faster than the non-expert trajectories.
The research team plans to demonstrate that the robot can learn to accurately and reliably perform representative surgical sub-tasks, ideally at speeds faster than an expert surgeon. The hypothesis is that the Raven tele-robotic system can reliably isolate, cut of circulation to, control, and excise on the physical phantoms of human pediatric appendices. In a series of experiments, the researchers will use the Raven’s sensors to record trajectories as these subtasks are repeatedly performed by human expert surgeons, process the data to generate optimal robot trajectories, and perform these trajectories and evaluate the results. The plan is to establish evaluation metrics and analyze results over multiple trials to draw conclusions about reliability to assess their Hypothesis.
Using an apprenticeship learning approach, the researchers can extract the intended trajectory from the human demonstrations and use it as the reference trajectory specifying the particular task, and applied it to knottying. Their experimental results suggest that they can use the learned trajectory model for the task to perform autonomous knot-ties from fairly well known initial conditions. The CITRIS seed grant will permit integration among the campus groups and experimental proof-of-concept with quantitative results that are necessary to apply for federal funding. The UC Santa Cruz surgical robotics platform (RAVEN) is one of the most advanced surgical robotics research platforms. The RAVEN is higher performance (bandwidth) and more reliable, the platforms are also more compact, enabling setting them up side by side which is how such robots are used in practice. This seed grant will permit the research team to apply apprenticeship learning algorithms on the Ravens and perform experiments with pioneering surgeon Doug Boyd from UC Davis.