Robotic systems have been widely employed for many needle biopsy applications, such as transthoracic lung biopsies that enable early diagnosis of primary lung cancers with small, distal lesions. The CRANE robot, a robotic needle positioning system for CT guided procedures, is one such system that has demonstrated a low-profile, high dexterity, and a large number of degrees of freedom. In this thesis, we present an end-to-end automated control scheme for the CRANE robotic platform that optimizes a robotic configuration over several defined metrics relevant to the successful completion of a transthoracic needle biopsy procedure. For a given needle tip target pose, multiple viable arm configurations are generated and ranked based several metric criteria. Once a configuration is chosen, a planner is run to find a collision-free path from the robot’s initial position outside of the CT-bore to this final configuration within the bore.

Fast and efficient path generation is critical for robots operating in complex environments. This motion planning problem is often performed in a robot's actuation or configuration space, where popular pathfinding methods such as A*, RRT*, get exponentially more computationally expensive to execute as the dimensionality increases or the spaces become more cluttered and complex. On the other hand, if one were to save the entire set of paths connecting all pair of locations in the configuration space a priori, one would run out of memory very quickly. In this work, we introduce a novel way of producing fast and optimal motion plans by using a stepping neural network approach, called OracleNet. OracleNet uses Recurrent Neural Networks to determine end-to-end trajectories in an iterative manner that implicitly generates optimal motion plans with minimal loss in performance in a compact form. The algorithm is straightforward in implementation while consistently generating near-optimal paths in a single, iterative, end-to-end roll-out. In practice, OracleNet generally has fixed-time execution regardless of the configuration space complexity while outperforming popular pathfinding algorithms in complex environments and higher dimensions.

Robots have played a crucial role in industries for decades, streamlining manufacturing processes and performing tasks with precision. However, with the availability of more affordable components and the rapid advancement of AI techniques, there is a growing interest in integrating robots into everyday settings. The challenge lies in the decades-old algorithms used in industrial robots often assume highly structured and controlled environments. These algorithms struggle to adapt to unstructured environments commonly found in everyday life, where obstacles and uncertainties abound. As a result, there is a need for newer, more adaptable algorithms to make robots a practical and safe part of our daily lives.

Motion planning, an age-old challenge in robotics, revolves around the task of determining a safe and efficient path for a robot to navigate an environment while evading obstacles. This problem is central in various robotics applications, from developing autonomous vehicles and healthcare robots to household robotic assistants. While existing methods have proven effective in generating trajectories and paths for pre-defined, structured environments, they face significant challenges when dealing with robots boasting higher degrees of freedom and adapting to changing environments. In these cases, the generated trajectories often necessitate further refinement before a robot can successfully execute them. In response to these limitations, there has been a recent surge of interest in using learning-based methods, which can address the shortcomings of traditional planners and enhance the generalizability and efficiency of robot motion planning algorithms.

This thesis introduces efficient planning algorithms that exhibit remarkable adaptability to various environments. Leveraging techniques from large language models, specifically the versatile Transformer architecture, we demonstrate how our planning algorithms can rapidly generate efficient trajectories while generalizing across diverse environmental contexts. Notably, we showcase the capacity of our learned models to tackle complex motion planning challenges, such as constraint planning, without the need for additional training data. By introducing a novel constraint function to encode the variabilities inherent in planning environments, we also lay the foundation for capturing and addressing different sources of uncertainties in the planning process. Looking ahead, we anticipate that our approaches will not only be readily accessible but also broadly beneficial, facilitating the seamless transfer of learned motion planners into a myriad of robotic-environment interaction scenarios and ushering in new possibilities for innovation and practical application in robotics.

Applications of rolling diaphragm transmissions for medical and teleoperated robotics are of great interest, due to the low friction of rolling diaphragms combined with the power density and stiffness of hydraulic transmissions. However, the stiffness enabling pressure preloads can form a tradeoff against bearing loading in some rolling diaphragm layouts, and transmission setup can be difficult. Utilization of cable drives compliment the rolling diaphragm transmission’s advantages, but maintaining cable tension is crucial for optimal and consistent performance.In this thesis, a coaxial opposed rolling diaphragm layout with cable drive and an electronic transmission control system are investigated, with a focus on system reliability and scalability. Mechanical features are proposed which enable force balancing, decoupling of transmission pressure from bearing loads, and maintenance of cable tension. Key considerations and procedures for automation of transmission setup, phasing, and operation are also presented. An analysis of system stiffness is presented to identify key compliance contributors, and experiments are conducted to validate prototype design performance.

In recent decades, advances in minimally invasive surgery has led to shorter patient recovery times, less pain, and less injury to healthy tissue. The introduction of robotic assisted minimally invasive surgical systems promises to improve surgeon precision and performance to further improve positive outcomes. The idea of dexterous operation within a lumen of the body for medical observation, biopsy, and therapeutics delivery can be enabled, in particular, by continuum robots. Capable of replicating locomotion, dexterity, and flexibility found in biology, from tentacles, vines, and snakes, continuum robots offer unique characteristics over rigid-link robots. These robots have shown clinical promise by entering the body through natural orifices and maneuvering inside complex anatomical structures to complete tasks. Although the technology is promising, the cost and bulky size of current systems ultimately limits broader access to current technology by patients and clinicians.

This thesis presents a novel, handheld, tendon-driven continuum robot catheter for augmentation of an endoscope. By using this steerable robotic catheter inside the endoscope’s working channel, the robot offers greater dexterity and instrument maneuverability within an anatomical workspace. The performance of the presented robotic catheter has been validated via benchtop experiments and multiple ex vivo experiments. Regular feedback from clinical collaborators drove a user-oriented design that is lightweight, easy to use, and does not significantly interfere with the familiarity clinicians have with endoscopes. In this work, the majority of volunteer clinicians were otolaryngologists, but the technology is applicable to other medical specialties and not limited to bronchoscopy endoscopes.

Surgical robots are becoming increasingly common in operating rooms, which provides the opportunity to deploy automation algorithms for surgery. Surgical task automation aims to improve patient throughput, reduce quality-of-care variance among surgeries, and potentially deliver complete automated surgery in the future. While progress in developing autonomous surgical tasks has leaped forward, reactive maneuvers to traumatic events, such as hemostasis, represent a critical area that has attracted little attention. Hemostasis describes a state of the surgical field that is achieved when there is no site of active bleeding and the tissues are unobstructed by blood. Unlike previously automated tasks that occur in a more predictable cadence within a procedure, bleeding can be unpredictable, which necessitates hemostatic maneuvers at any time during surgery.

In this dissertation, all the necessary perception, motion planning, and control strategies are presented to autonomously control a robotic suction tool to clear the surgical field from blood. First, a surgical tool tracking technique is proposed that localizes the robotic agent, which will clear the surgical field, in the endoscopic camera frame. The surgical tool tracking is combined with a deformable tissue tracker to completely track a surgical scene before a vessel rupture occurs. The combination of the two trackers is coined SuPer, the Surgical Perception framework. Next, the blood from a vessel rupture scenario is perceived by detecting and reconstructing the flowing blood from the endoscopic camera data. Finally, a controller and a motion planner for the robotic suction tool to clear the surgical field of blood are presented.

Motion planning is one of the most critical tasks in robotics, as it is one of the few critical functions for robot autonomy. This component requires fast computations and generalization to different environments for problems such as collision avoidance. Deep learning, as a new fast-growing field, offers great advances in computational speed and generalization. It has shown success in computer vision and reinforcement learning, which is closely related to motion planning. In this thesis, we will investigate the combination of learning and motion planning methods. Specifically, we separately consider individual components of motion planning tasks. By combining the proposed learning-based methods for each component, we can obtain an integrated end-to-end learning-based motion planning algorithm. We show experimental results for each component. In general, our learning-based methods showed high computational speed with generalization in several motion planning tasks.