Data-driven modelling and control of concentric tube robots with application in distal lung sampling
This research aims to explore the use of Concentric Tube Robots (CTRs) as a novel alternative to needle-based interventions in order to make these procedures more accurate and repeatable. CTRs due to their small footprint, compliance, and dexterity have been proposed for several minimally-invasive robotic surgeries. As a novel flexible robot, it has the potential to reach distal parts of the human lung that are difficult or impossible to reach with conventional needle-based interventions. There are, however, still significant challenges associated with the motion and position control of CTRs. Commonly used model-based control approaches are computationally expensive to solve and often employ simplified geometric/dynamic assumptions, which could be inaccurate in the presence of unmodelled disturbances and external interaction forces. Consequently, this work explores different control strategies to overcome these limitation. This is achieved by first building a simulation environment based on a computationally improved kinematic model that enables real-time control. Then, data-driven control approaches are investigated in order to provide accurate position control in the presence of uncertainties in the system. Finally, a three-phase affordance-aware motion planner is proposed to demonstrate the feasibility of using CTRs for percutaneous lung biopsy. According to this, the first part of this work concentrates on computationally efficient real-time modelling and simulation of CTRs. In order to achieve this, two approaches are taken. The first one introduces a method that can rapidly estimate the solution of the kinematic model, while the second approach focuses on implementing the existing model in a computationally efficient way in Robot Operating System (ROS) using C++. Second, this work explores data-driven solutions to control the robot without relying on the kinematic model. Consequently, two data-driven solutions are proposed, namely the Hybrid Dual Jacobian approach and the Extended Dynamic Mode Decomposition (EDMD) algorithm. The hybrid controller combines the advantages of model-based and data-driven control approaches, while the EDMD provides a completely model-free solution to control the robot. Both controllers are capable of rapidly predicting the robot’s nonlinear dynamics from a limited data set and offer consistent control under external loading and in the presence of obstacles. The third part of the thesis explores the use of CTRs in the context of distal lung sampling. This work demonstrates that CTRs are suitable for Needle-Based Optical Endomicroscopy where a CTR steers a fluorescent imaging probe with cellular and bacterial imaging capability inside soft tissue. Then, it is also demonstrated that a CTR can be used as a Steerable Needle to reach a target deep inside the tissue. To achieve these tasks, a motion planner is essential due to the fact that a CTR is only capable of reaching specific points in its workspace and there are a number of configurations where the robot becomes unstable. Based on this, a threeii phase affordance-aware motion planner algorithm is developed. The motion planner selects the best entry point for a specific task. Based on the selected entry point it first generates a stable trajectory from the robot’s initial configuration to the selected entry point. Then, a feasible trajectory is generated from the entry point to the target. Finally, the proposed datadriven control algorithm is applied to autonomously steer the robot on the generated trajectory toward the target region for endomicroscopic imaging.