Control Architectures



In the last decade very complex systems have emerged, especially in the area of service and assistance robotics. Some are even available as products. This became possible due to more powerful computer systems, better sensors, new methods of perception, innovative methods of localization and mapping, as well as cognitive system components. This positive trend was boosted by the increasingly extensive open source robotics libraries. They enable a much faster the construction of basic systems. Nevertheless, the growing complexity of robotic systems remains a major challenge.

The Robotics Research Lab has been researching the development of efficient robot control architectures more than a decade. This includes the selection of suitable embedded computer architectures, the support for control software development by appropriate robotic frameworks, concepts for control architectures as well as methods for system validation and verification (see figure). The demands from application side on system characteristics such as adaptability, real-time capability, reliability, scalability and security must be ensured.


To provide the above mentioned support for the system designer, the research group has been investigating the design of robotic frameworks. The current development FINROC - Framework for Intelligent Robot Control (see - offers a variety of tools to easily integrate simulation components or open source libraries, to provide the user with generic GUIs and graphical programming tools, and to allow a monitoring system at runtime. Due to its modular and compact structure, FINROC can also be deployed on smaller computer nodes in principle.

Hence a current goal is the extension of Finroc to embedded systems such as microcontrollers and FPGAs. The Goal is hereby the development of distributed and scalable systems without introducing a major break in the development flow.

Another thread of research activities evolves around the behavior-based architecture iB2C. The basic idea of this architecture is that a complex system behavior emerges from the fusion of simple behavior nodes. It was shown that this architecture can perfectly adapt to sensor noise, environmental characteristics, or to changes in robot characteristics. In order to master the control architectures, consisting of hundreds of nodes, design tools have been developed,  automated methods for detecting cycles have been examined, and adaptive, reliable robot control architectures and approaches to formal verification of system parameters have been tested. The validation and verification of the behavior of networks will be studied in particular to guarantee security and reliability for different robot systems. Another aspect is the automatic generation of the behavior parameters. They are critical to the performance of the behavior-based control system. In future these parameters should be automatically determined by machine-learning methods.