This Small Business Innovation Research (SBIR) Phase II research project will create an information-based robotic grasping framework to enable practical grasping of objects for any robotic manipulator and any robotic hand, or even multiple hands. Grasp algorithms are stored in an XML database organized in a tree structure that allows rapid access and uses intelligent caching for very large databases. When a new object is presented to the grasping system, best matches are found in the database and the corresponding algorithms are extrapolated to determine the best grasp for the new object. Shape, surface properties, and articulation are used for matching. The techniques support the grasping of moving objects that can be tracked with a vision-based system. For constructing the grasp database, human supervisors train new grasps by simply picking up objects and giving special cues. Collection devices, such as data gloves and machine vision systems, are used to collect the supervisor?s hand position and contact forces, and a learning module finds new grasps by coupling supervisory input with simulation-based optimization, using high-fidelity dynamic modeling. For optimization, control and configuration parameters (in end-effector space) are perturbed iteratively using nonlinear numerical optimization techniques.
If successful the creation of a comprehensive grasping framework as proposed in this project will have broad impact to research, industry, and society. Traditional grasping systems require specialized coding for new tasks and new robots. The proposed system will facilitate specific instantiations of general grasping algorithms. Application to virtually any robot manipulator, any hand, and any object to be grasped will be possible. This unprecedented flexibility, coupled with advanced and innovative grasping algorithms will play a role in advancing general purpose robots (those that can do multiple tasks without reprogramming). Robots with the ability to grasp hold promise for industries with labor shortages. The agricultural industry, for instance, will use robotic grasping for harvesting. Grasping robots will work in dangerous environments. An example application is rescuing injured humans in dangerous situations. Next-generation robots will assist the disabled with intelligent manipulators that can open doors and pick up objects. Grasping robots will support manufacturing and warehouse businesses. The simulation capability that is part of this research will allow new grasping strategies to be tested safely in a virtual environment before being implemented and fielded.
Phase II Energid created an information-based robotic grasping and manipulation framework to enable practical application of any robotic manipulator to solve real-world problems. With the new system, grasping and manipulation algorithms are stored in an XML database organized in a tree structure that allows rapid access and uses intelligent caching for very large databases. Behaviors can be easily specified through a tree structure. When a new object or manipulation scenario is presented to the grasping system, matches are found in the database and the corresponding algorithms are extrapolated to determine the best manipulation strategy. Shape, surface properties, and articulation are used for configuring the strategy. The technology enabling this was broad, including robot control, machine vision, dynamic and kinematic simulation, and user-interface components. Energid has applied the results of the research to the development of a system for underwater oil drilling, the development of an inspection robot for the nuclear industry, the control of a cleaning robot, control of an airport security robot, control of a humanoid robot for entertainment, control of a modular robot system, a medical robot, and a large number of research and educational applications. Energid has licensed and extended the software for manufacturing inspection to one of the world's largest contract manufacturers, Jabil Circuit. Phase IIB Energid integrated feature- and subspace-based visual servoing algorithms into robot control; applied accelerometers and other sensors for robot sag compensation; attached 3D mouse for easier interactive robot control; developed algorithms for both robot pose calibration and vibration control; implemented methods to improve accuracy; designed algorithms for leveraging CAD models for control and inspection; and developed a graphical user interface for work flow visualization and task assignments. This work represented important advancement in the science of robot control and its incorporation into our Actin software toolkit makes it practical and applicable immediately to robots under development and in the field. This work enables the application of sensors to improve the performance of robot arms. Traditional industrial robots are expensive, rigid, heavy and dangerous. While these robots have their place in industrial settings, they are limited in two profound ways: 1) it is hard to reduce their cost and 2) it is not possible to install them safely next to a person. As of the writing of this report, we are using revenues from products developed using the new technology, enhancing algorithms and integrating sensors to enable lower cost robots to perform useful tasks. These lower cost robots are not as rigid or as accurate mechanically, as traditional industrial robots, and they instead rely on sensors and algorithms to perform tasks. In particular, the capabilities developed during the Phase IIB project have let us take the Robai Cyton robot product from an arm targeted for R&D to one, the Cyton Gamma R3, suitable for an initial set of automation tasks. Energid has raised additional funding explicitly to build on our NSF effort and advance the Cyton. Marketing Activities Energid had a booth in the Eureka park section of the CES 2013 to promote our robotics software and the Cyton arms. Energid also attended Automate 2013 and RoboBusiness 2013. Industry and Service Applications: The capabilities developed during the NSF Phase II project have enabled the Cyton robot product from an arm targeted for R&D to one suitable for an initial set of automation tasks. We have worked with Intel, Jabil, WaveConnex and Flextronics for industrial applications; and Johns Hopkins University and Robot Rose for service robot applications. Investments In June, 2013, MassVentures awarded Energid a $100k START grant based on the NSF SBIR. MassVentures, with seasoned venture investors and people with substantial industrial automation experience, is excited about the market potential for the Cytons. They accelerated the START grant and want to join the Series A round in Q2 2014. Other investment opportunities include Mitsui, Intel capital, and Flextronics.
