A major goal in robotics is the successful demonstration of mobile, autonomous robots. The computational power needed for sensory processing, planning, learning, and motion control is staggering. Some of this need can be met by using special purpose devices, but much computation is still needed for the higher level functions. A computer architecture for controlling an autonomous, mobile robot must be flexible, have very high performance, and support real-time execution. Execution of the tasks can be organized into a hierarchy. This project employs a hierarchical, MIMD parallel architecture for this application. The Hierarchical MultiProcessor consists of several levels; each level is a group of microprocessors connected by two buses. Communication between processors is accomplished by message passing. One of the buses at each level is also used for communication between adjacent levels. Several novel techniques of improving the communication efficiency in this architecture are presented. A major thrust in this project is to perform a detailed performance analysis of the architecture and its variations. The complexity of the multiprocessor, multi-bus architecture, and of the real-time task execution, presents formidable challenges for performance analysis. Techniques for doing this analysis, are being developed. The Robotics and Intelligent Systems Laboratory at N.C. State is in the process of building a mobile, autonomous robot. A reduced version of the HMP architecture is being implemented, which will be followed by a full-scale implementation.
