Use of a Multiprocessor for Control of a Robotic System

Abstract

In the control of robotic systems, multiprocessors have a number of potential advantages over single-processor configurations. These advantages include the possibility of modular future expansion, increased speed, and fault tolerance. This paper describes both the hardware and software design of a multiprocessor system and its use in controlling the OSU Hexapod vehicle, a complex robotic system. With the design of an inexpensive, high-density, eight-channel parallel line unit (PLU), it is possible to link five LSI-11 processors in a fully connected configuration. This configuration is very valuable for experimentation, since trials of tree, star, and loop structures can be made by simply ignoring links that are not needed. First, a tree structure is used to implement a hierarchical control algorithm run on a previously used uniprocessor system. Next, fault-tolerant features are added so that failed units are automatically replaced by spare units or, if all spares are in use, the necessary tasks are covered by the central processor, with a decrease in servoing frequency. To decentralize control to improve overall reliability, a ring structure is also used. If a processor fails, the ring is automatically reconfigured to isolate the faulty unit. Another application for the multiprocessor in the control of the OSU Hexapod vehicle is for real-time optimization of leg tip forces. This task demonstrates the usefulness of the multiprocessor in a task that is not strictly parallel in form.