Modeling by Manipulation

Trustees Of Boston University, Boston

Investigators

Abstract

In this project, perceptual systems will be developed for modeling objects as they are manipulated during teleoperation. These systems will estimate those object properties that mediate the interaction between the robot and its environment, including object geometry, mass, moment of inertia, and friction. The inputs to the system will be sensor data streams from the remote environment, including force, position, and video signals, and the outputs will be models of the manipulated objects and their properties. This approach, modeling by manipulation, will create new robot perceptual capabilities in unstructured environments. In doing so, it will address a fundamental challenge in the development of autonomous robots: the ability to sense and model the environment. The proposed system will be built in the context of teleoperation to take advantage of the human operator's motor skills to execute the manipulation tasks. The sensory information returned from the remote robot will provide a comprehensive description of the task. By focusing the project's theoretical developments on specific test beds, tasks and properties, the essential perceptual capabilities will be developed to enable new progress on the other components of autonomy, including planning and control.