TRACK II: A multi-processor robot controller

Abstract

Advanced robotic systems require powerful controllers which can perform complex tasks involving multiple robots and sensors, while meeting strict safety requirements. These controllers require a multi-processor architecture, a real-time operating system and appropriate interfaces. Most robot controllers are based on bus architectures and asynchronous multi-tasking operating systems. Such configurations involve significant overheads for inter-task communications and for task switching, that can degrade performance as the system is extended. The asynchronous nature of these systems causes non-deterministic behavior and poor reliability. This paper presents the specifcation and design of a high performance controller for advanced manipulation systems. The controller has a modular parallel architecture which can be extended without performance degradation, a synchronous real-time operating system and a kit of hardware modules. Our design philosophy aims for maximum real time computing power by exploiting the computational characteristics of robot control and planning algorithms, together with minimum sojiware development effort, at the expense of cpu utilization.