Robust Intelligent Manipulation and Apprenticeship Learning for Robotic Surgical

Principal Investigator: Ron Alterovitz
Funding Agency: Case Western Reserve (NSF)
Agency Number: RES504243

Abstract
Robotic surgical assistants (RSAs) such as Intuitive Surgical’s da Vinci system are increasingly being accepted in hospitals for endoscopic surgery. Dual manipulator RSAs are usually tele-operated in a master-slave mode. As a result many surgical tasks remain extremely tedious and time consuming.

To assist surgeons, the PIs and our students propose to investigate new models, algorithms, and testbeds for robust intelligent manipulation that will permit supervisory control for RSAs. Consider two scenarios: 1) one RSA manipulator is activated in an autonomous mode to retract and hold tissue at a given tension to assist the surgeon during dissection, and 2) data from observation of motions and video of human experts is processed to allow an RSA to autonomously suture surgeonspecified tissues. Achieving these and related goals will require a collaborative new research effort.

In this medium-scale project, we will investigate both analytic and empirical approaches to robust intelligence in the context of minimally invasive surgery drawing on the combined expertise of the PIs. Analytic approaches require us to define mathematical models of system state, deformable tissue response, kinematics, dynamics, motion planning, and explicit uncertainty models. Empirical approaches require us to develop new performance metrics and stochastic machine learning techniques. To evaluate, understand, and compare analytic and empirical approaches and new hybrid approaches, we will develop simulation and hardware testbeds.

Intellectual merit: This project will advance basic scientific understanding by developing and evaluating new analytic models and algorithms for robust grasping and manipulation of deformable tissues, which will extend and integrate our preliminary investigation of Natural Admittance Control, Deformation-Space, Stochastic Roadmaps for Motion Planning, and Modeling/Simulation of Deformable Objects. The project will also develop new empirical approaches based on our work on Apprenticeship Learning, Dynamic Time Warping, and Expectation Maximization for control trajectory optimization. The project will extend, compare, and evaluate these methods and new methods, such as, Simultaneous Modeling and Planning, within the focused context of providing
robust intelligence for endoscopic robot surgical assistants (RSAs).

Broader Impacts: Robust intelligent manipulation for RSAs has the potential to significantly improve patient health by ensuring precision and accuracy, thereby speeding patient recovery time and minimizing side effects. Robust intelligence for RSAs also has the potential to enhance surgeon performance when expert assistants are not available, reduce medical errors by lessening surgeon tedium and fatigue, and cut costs by decreasing time in the operating room. The developed robotic motion control, planning, and learning algorithms will also be directly applicable to the robust intelligent manipulation of deformable objects in non-surgical environments, including robotic assembly of composite materials, manufacturing of textiles, and automated handling of food products.

We will integrate this research with education including outreach activities with local high schools, actively involving undergraduates in the research activities, and by introducing lecture material and projects into graduate courses. Special emphasis will be given to recruit qualified students from underrepresented groups, including minorities and women engineers, in collaboration with the minority student offices at the Case School of Engineering, UNC Department of Computer Science, and UC Berkeley College of Engineering