Computer vision for guiding manual assembly

Abstract

This paper considers an augmented reality interface for guiding manual assembly. The key challenges involved in enabling such a system include the design of an effective information presentation scheme, as well as robust and fast sensing of the assembly scene. We present a framework for assembly scene augmentation that is based on robotic assembly planning. We then discuss the problem of sensing for augmented reality without the use of fiducial and present our formulation that combines model-based and appearance-based approaches for recognizing assembly states.