The design and analysis of a computer system for sensory-based robotics

Many of today\u27s robots are used to perform simple and repetitive tasks using little sensory information. Such tasks are simple for they are carried out in tightly constrained environments in which all external uncontrollable forces have been removed. In the near future, robots are going to be called upon to perform more complex tasks in relatively unconstrained environments. For example, current plans call for autonomous robots to be used for the construction and maintenance of NASA\u27s space station. Such robots will be more complex, dextrous, sensory driven, and will require more sophisticated control systems. In this dissertation, we present the design and analysis of a new control system, called the RS control system (Robot Schema Control System), that will be used for controlling these complex, sensory driven robots. The design of the RS control system has been driven by the computational characteristics of the robotics environment. In other words, we have analyzed the computational power that will be required to control robots in unconstrained, dynamic environments and have used the results from this analysis in designing the basic features of the RS control system. The RS control system is composed of three major components: (1) a shared memory multiprocessor architecture, (2) a dynamic real-time operating system, and (3) a real-time programming language. This dissertation primarily focuses on the dynamic real-time operating system. The major component of the operating system is the Dynamic Multi-Pool DMP scheduler whose primary function is the dynamic assignment of processes to processors as processes become ready to execute. The structure and functionality of the DMP scheduler, as well as results obtained using the DMP scheduler in a simulated multiprocessor environment, are presented. We have been able to show that it is feasible (by using the DMP scheduler) to construct multiprocessor systems composed of large numbers of processors and to efficiently use such systems in the control of complex robots. These results can now be used in the construction of the RS control system