Increasing the level of autonomy in robot-assisted surgery has the potential to improve the safety, speed, and applicability of robot-assisted surgical systems. To facilitate the development and incorporation of robot autonomy in clinical settings, human-robot collaboration models have been suggested in which human and robotic agents work together to accomplish a task. In this work, we measure performance of several human-robot collaboration models in two experiments based on the task of segmenting a stiff inclusion in soft tissue, which simulates a tumor. In the inclusion segmentation experiment, twelve participants explored an artificial tissue and identified the inclusion boundary under the collaboration models of (1) teleoperation, (2) supervised control, (3) traded control, and (4) full autonomy. In the boundary identification experiment, we isolate the performance of human and robotic agents in the boundary identification sub-task; participants and a robotic agent independently identified the boundary of four virtually palpated tissues. Results from the inclusion segmentation experiment indicate that human agents complete the task faster; teleoperation had the fastest task times. Results of both experiments indicate that the robotic agent identifies boundaries with higher sensitivity and less variance than human agents. This indicates that task accuracy increases when a robotic agent segments the boundary, while including a human agent can decrease the overall task time.

In robot-assisted surgery, exploration and manipulation tasks can be achieved through collaboration among robotic and human agents. Collaboration models can potentially include multiple agents working towards a shared objective-a scenario referred to as multilateral manipulation. We present a flexible software framework to expedite development of various multilateral manipulation strategies. We demonstrate the effectiveness of an implementation of the framework in a palpation task. Five different collaboration models were tested in which the goal of the multilateral manipulation system is to segment a stiff inclusion from its surrounding soft tissue: three of these collaboration models used machine learning methods for segmentation, and two required human operator segmentation. The collaboration models tested were: 1) fully autonomous exploration of the tissue; 2) shared control between a human and robotic agent; 3) supervised control where the operator dictates commands to the robot; 4) traded control between the two agents; and 5) bilateral teleoperation. Results indicate tradeoffs in sensitivity, maximum force applied, safety implications, and duration of experiment among the five models.

In robot-assisted surgery, manipulation tasks can be achieved through collaboration among robotic and human agents. Collaboration models can potentially include multiple agents working towards a shared objective - a scenario referred to as multilateral manipulation. In this work, we examine multilateral manipulation in the task of debridement: removing dead tissue or damaged tissue fragments to allow remaining healthy tissue to heal. We extended our previously developed multilateral manipulation software framework to the task of debridement and implemented four different collaboration models: (1) fully autonomous debridement, (2) shared control between a human and robotic agent, (3) supervised control where the operator identifies foreign bodies to be excised, and (4) teleoperation. We demonstrate these collaboration models with the RAVEN-II robot, an open-architecture surgical robot with two cable-driven 7-DOF arms. Each collaboration model included methods to remove foreign bodies from the field of view of a stereoscopic camera setup. Results indicate tradeoffs in experiment duration, hardware requirements, and safety implications among the four collaboration models.

Localizing tumors and measuring tissue mechanical properties can aid in surgical planning and evaluating the progression of disease. In this paper, autonomous robotic palpation with supervised machine learning algorithms enables mechanical localization and segmentation of stiff inclusions in artificial tissue. Elastography generates training data for the learning algorithms, providing a noninvasive, inclusion-specific characterization of tissue mechanics. Once an embedded hard inclusion was identified in the elastographic image, Gaussian discriminant analysis generated a classifier to threshold stiffness values acquired from autonomous robotic palpation. This classifier was later used to classify newly acquired points as either part of the inclusion or surrounding soft tissue. An expectation-maximization algorithm with underlying Markov random fields improved this initial classifier over successive iterations to better approximate the boundary of the inclusion. Results demonstrate robustness with respect to inclusion shape, size, and the initial classifier value. For three trials segmenting a cubic inclusion, sensitivity was above 0.95 and specificity was above 0.92.

Localizing tumors and measuring tissue mechanical properties can be useful for surgical planning and evaluating progression of disease. In this paper, supervised machine learning algorithms enable mechanical localization of stiff inclusions in artificial tissue after autonomous robotic palpation. Elastography is used to generate training data for the learning algorithms, providing a non-invasive, inclusion-specific characterization of tissue biomechanics. In particular, elastography was used to characterize the stiffness of artificial tissue with an embedded hard inclusion. Once the inclusion was identified on the elastographic image, machine learning methods identified the difference in stiffness between the inclusion and surrounding soft tissue and generated classifiers, which were used to label stiffness values as either part of the inclusion or soft tissue. Next, data acquired via autonomous robotic palpation of the artificial tissue created a map of the stiffness distributed over the surface of the tissue. The points in this map were thresholded against the classifiers trained by the machine learning algorithms, and points theorized to belong to the hard inclusion were labeled. Centroid approximations of the hard inclusion based on this labeling show that classifying stiffness data acquired by autonomous robotic palpation and labeled by a classifier trained from elastography data provides a more accurate method of localizing hard inclusions than using unclassified data.

This paper reports the results of an in-water laboratory evaluation of the state estimation accuracy of a commercially available inertial navigation system (INS), and the effect of variations in the accuracy and the update rate of INS correction data on the accuracy of the INS state estimate. The navigation accuracy of the INS state estimate is evaluated by comparison with navigation data from a highly accurate navigation system available in our laboratory that employs high-frequency acoustic time-of-flight ranging, precision pressure depth sensing, bottom-lock Doppler sonar, and a true-North-seeking gyrocompass that is integral to the INS.