Autonomous systems demand new computer system architectures and new development strategies

This paper addresses two related problem areas with regard to autonomous systems, namely (i) appropriate architectures of the embedded computing systems, and (ii) appropriate system development strategies and tools. Adaptivity and learning are identified as important characteristics of autonomous systems. A modular architecture capable of executing multiple-neural-network systems with real-time learning and control abilities is described. This architecture is based on distributed nodes of massively parallel processing elements communicating by means of a scalable high bandwidth medium. The nodes and the medium operate in a time-deterministic manner. Further, it is argued that autonomous systems must be developed through a process of system alteration-change-orientation rather than the phase-oriented specify-design-implement-and-test model which is common today. A tool which, in cooperation with the system architecture, supports this paradigm is described. Required functions of the development tool are, e.g., inspection and visualization of data, change of software and hardware functions, and ability to cope with changes in the mechanical structure, including sensors and actuators, all while the autonomous system is in action. A graphical, interactive design environment that meets these demands is outlined.<>