Affordance-Based Control of a Variable-Autonomy Telerobot

Most robot platforms operate in one of two modes: full autonomy, usually in the lab; or lowlevel teleoperation, usually in the field. Full autonomy is currently realizable only in narrow domains of robotics—like mapping an environment. Tedious teleoperation/joystick control is typical in military applications, like complex manipulation and navigation with bomb-disposal robots. This thesis describes a robot “surrogate” with an intermediate and variable level of autonomy. The robot surrogate accomplishes manipulation tasks by taking guidance and planning suggestions from a human “supervisor.” The surrogate does not engage in high-level reasoning, but only in intermediate-level planning and low-level control. The human supervisor supplies the high-level reasoning and some intermediate control—leaving execution details for the surrogate. The supervisor supplies world knowledge and planning suggestions by “drawing” on a 3D view of the world constructed from sensor data. The surrogate conveys its own model of the world to the supervisor, to enable mental-model sharing between supervisor and surrogate. The contributions of this thesis include: (1) A novel partitioning of the manipulation task load between supervisor and surrogate, which side-steps problems in autonomous robotics by replacing them with problems in interfaces, perception, planning, control, and human-robot trust; and (2) The algorithms and software designed and built for mental model-sharing and supervisor-assisted manipulation. Using this system, we are able to command the PR2 to manipulate simple objects incorporating either a single revolute or prismatic joint. Thesis Supervisor: Professor Seth Teller Title: Professor of Computer Science and Engineering