Abstract
Skilled labor shortage and a competitive market are challenges for many businesses. Hence, automation and robotization are well-established in the large-scale, well-structured industrial environments of many factories. Moreover, there is a push for safer and more easily programmable robots, so-called cobots, that would be more suitable for smaller-scale applications. However, the implementation and customization of complex robot systems with different sensors, tools, external systems, etc. in particular for the use cases of small and medium-sized enterprises (SMEs) are still often infeasible. Therefore, this paper describes the design, implementation, and testing of a flexible mobile manipulator architecture with the following three abstraction layers. A high-level semantic layer models and utilizes OWL ontologies of abstract manufacturing processes, specific workcell environments, linked context knowledge, and the current semantic world state. An intermediate translation layer collects and exchanges data from multiple system components via a unified internal database, which is used, e.g., in combination with behavior trees to convert symbolic, high-level actions into multiple parameter-driven low-level commands. A low-level control layer is implemented locally on a mobile robot to provide a unified interface of its potentially customized subsystems to the higher layers. This architectural approach facilitates the implementation of new use cases, e.g., by a system integrator, via flexibly adapting semantic representations and behavior trees on the model level without changing source code. The implemented system was deployed in five different real-world use cases of industrial partners in the VOJEXT project. This paper focuses on the use case of plasterboard wall preparation from the construction domain, which includes taping, spraying, and sanding operations, to evaluate the particular requirements of this application and how they can be met by the proposed robot control architecture.