Easy-to-use, Autonomous Bin-picking and Assembly Operations for the Manufacturing Industry

We will develop a series of object manipulation primitives to pick up and assemble standard mechanical parts such as screws, gears and pulleys that can be configured without any programming skills. Building up on a smart robotic gripper, 3D perception and machine learning algorithms, we will design a graphical user interface that allows an user to label arbitrary 3D objects, identify assembly points, and issue pick-up and assembly commands, which are modeled by standard industrial assembly tasks that have been identified by NIST’s manufacturing group. Specifically, we will provide routines for peg-in-hole and hole-on-peg assemblies that are defined by circular or rectangular holes or extrusions, as well as pick-up routines for arbitrary objects from clutter or in bins. The underlying algorithms have already been validated, making the focus of this proposal to develop interfaces that make their configuration as simple as possible, only requiring a user to present an individual object, labeling it, and selecting a desired action. This Phase II proposal will extend this approach to a larger class of objects and tasks and lead to deployment with real users.