Agile robots like the DLR Two-Hand-Arm-System "Justin" (with 43 DOFs, >250 sensors and >1kHz control rate over all DOF) demand for scalable, distributed computing resources and fast communication in hard realtime, working at the limits of current hardware capabilities. On the other hand, agile robots are complex systems built by a heterogenous team of tightly interacting experts, demanding for an agile development flow, that is: flexible, iterative and rapid. Crucial for this is to allow for a functional view on the system architecture decoupled from hardware details.
The aRD software concept was specificaally designed to balance these two contradicting requirements by providing an abstract view of a robotic system as a "net of functional blocks and communication links" and tools that add only a thin layer above the (realtime) OS to map the net of blocks onto distributed computing hardware.
In this talk we discuss design considerations and the current implementation of the aRD concept. We also demonstrate the system's performance with applications ranging from test stands up to full systems like Justin.