Projecting Product-Aware Cues as Assembly Intentions for Human-Robot Collaboration

Abstract

Abstract Collaborative environments between humans and robots are often characterized by simultaneous tasks carried out in close proximity. Recognizing robot intent in such circumstances can be crucial for operator safety and cannot be determined from robot motion alone. Projecting robot intentions on the product or the part the operator is collaborating on has the advantage that it is in the operator’s field of view and has the operator’s undivided attention. However, intention projection methods in literature use manual techniques for this purpose which can be prohibitively time consuming and unscalable to different part geometries. This problem is only more relevant in today’s manufacturing scenario that is characterized by part variety and volume. To this end, this study proposes (oriented) bounding boxes as a generalizable information construct for projecting assembly intentions that is capable of coping with different part geometries. The approach makes use of a digital thread framework for on-demand, run-time computation and retrieval of these bounding boxes from product CAD models and does so automatically without human intervention. A case-study with a real diesel engine assembly informs appreciable results and preliminary observations are discussed before presenting future directions for research.