Towards Robot Systems Architecture

Just as special purpose computers and mainframes grew into the generalpurpose personal computers we use everyday, special purpose industrialrobots are evolving into more general purpose personal robots. Asrobots become more capable and universal, their applications are lesswell-defined or even unknown at design time. We will have to designrobots for classes of tasks rather than specific applications. Havingguidelines for how to best organize, interface, and implement robotsystems and reason about trade-offs, as we do in computerarchitecture, will become crucial for success. In this paper weintroduce and adapt some useful notions and principles from computerarchitecture to robot systems architecture. We argue that notions suchas locality of reference, balanced architectures, and boundedness (interms of IO, memory, and CPU) can be leveraged in robot systemsdesign, and in particular, in the design of distributed robot systems.