Apprenticeship Scheduling for Human-Robot Teams in Manufacturing

Traditional uses of robotic technology in manufacturing have seen robotic work physically caged-off and separated from human workers. Automotive manufacturing is a success case for manufacturing automation. However, the human element is still an incredibly important facet of automotive manufacturing. Manual work still counts for 50% of the build process and requires two-thirds of the factory footprint. The process to build a car remains a manual task, which requires the dexterous skills of people. Recent advances in robotic technology are opening up the possibility of integrating mobile robotic assistants into areas of the factory that have traditionally been reserved for human workers. Research in localization and mapping, computer vision, control systems, and learning-from-demonstration is allowing these robots to support non-value added tasks (e.g., fetching of parts) and even value-added tasks in some cases (e.g., welding, drilling, etc.). Successfully introducing this advanced robotic technology to work on human-robot teams requires an algorithm that can both safely choreograph the activity of these teams in real-time and generate robotic behavior that gains the trust of the human workers and managers. Human workers often develop a sense of identity, security, and purpose from their jobs or roles in factories. Similarly, human workers develop mental models or expectations of how the team as a whole should work together. These patterns and expectation of workflow can change from team to team even when working on the same job. Thus, giving robots the ability to understand how to complete the set of tasks in a way that is intuitive to their human teammates is paramount to the success of integrating robots into human workspaces. Developing an user-interface by which humans can command or directly operate robotic assistants also presents a challenge to the success of integrating robots into the human workspace. The human-robot interface has long been seen as a major bottleneck for the success of these robotic systems (Parasuraman, Mustapha, and Hilburn 1999), even though much effort has been conducted to improve the fluidity of the interaction (Clare et al. 2012). Other researchers have taken the tack of explicitly formulating the