This project seeks to make the supervision of robotic systems operating in complex domains similar to that of humans so as to increase productivity and capabilities. The software framework resulting from this project frees supervisors from the burden of unnatural low-level commands and instead allows them to describe tasks in a structured language that has the ability to express global and local objectives across time spans. The framework then automatically computes the necessary motions to enable the robotic system accomplish the assigned tasks. This necessitates a comprehensive treatment of planning to account for sophisticated tasks, robot dynamics, collision avoidance, robust replanning, and interactions with human supervisors. To addresses such complexity, the framework employs a novel probabilistic search with discrete abstractions and enhanced sampling capability to focus the search on the space of feasible motions that enable the robotic system to make progress toward accomplishing the assigned task. The framework also provides critical feedback information about the progress made to help supervisors adapt the specifications in response to challenges encountered during planning and execution. This project is expected to establish a new paradigm for supervised autonomy and impact the development of research and commercial software for robotic systems. Doing so has the potential to enhance applications of robotic systems such as underwater vehicles in surveying marine wildlife, inspecting harbors and offshore platforms.
