Improving inclusion segmentation task performance through human-intent based human-robot collaboration

Abstract

Introducing autonomy into robotic minimally-invasive surgery has the capability to improve surgical performance. While pure autonomy is not yet feasible for many surgical procedures, human-robot collaboration models, which operate on a combination of human and autonomous input, have the potential to foster the transition to autonomy and offer enhanced performance over pure human control. Building on our previous work on human-robot collaboration models, we focus on designing and testing collaborative control models based on human intent for an inclusion segmentation task. We seek to use learning techniques and support vector machine classifiers to predict both local regions of human interest and the level of interest in these regions; these predictions will then inform the role of the robotic agent in the task. The intent-based collaboration models will be tested in a human subject study for the inclusion segmentation task against teleoperation and pure autonomy to assess task performance. We anticipate that these collaboration models will outperform both pure human and pure autonomous control.