Taking control: Modular and adaptive robotics process control systems

Robotics systems usually comprise sophisticated sensor and actuator systems with no less complex control applications. These systems are subject to frequent modifications and extensions and have to adapt to their environment. While automation systems are tailored to particular production processes, autonomous vehicles must adaptively switch their sensors and controllers depending on environmental conditions. However, when designing and implementing the process control system, traditional control theory focuses on the control problem at hand without having this variability in mind. Thus, the resulting models and implementation artefacts are monolithic, additionally complicating the real-time system design. In this paper, we present a modularisation approach for the design of robotics process control systems, which not only aims for variability at design-time but also for adaptivity at run-time. Our approach is based on a layered control architecture, which includes an explicit interface between the two domains involved: control engineering and computer science. Our architecture provides separation of concerns in terms of independent building blocks and data flows. For example, the replacement of a sensor no longer involves the tedious modification of downstream filters and controllers. Likewise, the error-prone mapping of high-level application behaviour to the process control system can be omitted. We validated our approach by the example of an autonomous vehicle use case. Our experimental results demonstrate ease of use and the capability to maintain quality of control on par with the original monolithic design.