Modeling and Control of a Continuum Robot for Medical Interventions

Abstract

A challenge in vascular surgeries is the precise positioning of catheters. Robotic systems, especially flexible soft/serial-link robots, might be employed to assist surgeons during catheterizations by providing better hand-eye coordination and more reliable dexterity, especially in deep-seated regions within the human body. This dissertation aims to advance the modeling and control of catheterizations coupled with magnetic actuation. Although magnetically-actuated catheters/guidewires are structurally simpler than comparable devices, such magnetic tools exhibit complex mechanical behavior. This thesis focuses on how these dynamics can be analyzed in potentially clinically-relevant applications. The contributions are twofold and are presented in the form of two research questions. This dissertation suggests approaches to i) modeling the behavior of continuum manipulators and ii) devising control algorithms to enable their steering. A novel approach is presented to derive models for complex mechanisms with large deformations that respects their intrinsic properties and preserves the structure of the configuration space. A balance between computational bandwidths and prediction quality is established with a Neural Networks-driven model which can be used for real-time applications. Furthermore, the dissertation explores the use of Reinforcement Learning techniques, Attractor Dynamics approach, and the Execution Extended Rapidly-Exploring Random Trees method for an untethered magnetically actuated milliscale particle.