Blackboard based hierarchical intelligent control system

Abstract

This paper describes ongoing research dealing with a new rcpresentation of generic hierarchical control systcms. The strategy is to generalize the model following paradigm and standard control system architecture, together with a recursive definition of hicrarchy, to represent goal driven controllers. The suggested hierarchical representation, which is based on the definition of a canonical representation of each of the levels, allows for control based on symbolic data rcprescntations. The structure of the hierarchy is illustrated with a simple mobile robot control example. I n t r o d u c t i o n Rcscarch in autonomous multivariate control of large scale systems typically addresses hierarchical representation and control [1,2,3]. Whiie this is an appealing concept it has not been defined carefully enough to analyze or systematically apply. This paper proposes a specific model of hierarchical control, based analogically on traditional control theory. We report on work currently in progress addressing these issues. The mapping of the proposed hierarchical control systems into a blackboard implementation architecture is shown. The proposed model clarifies a number of issues rclatcd to model following, planning, and levels of abstraction. Issues of parallel and sequential processing are also discussed. Problcrn Formulation The motivation for hierarchical control is that given a dynamic goal specification, it is not necessarily compatible with the plant dynamics. Also, while a large control system could be implemented as a single level controller, the computational limitations make many applications unfeasible. Hierarchical control addresses thEsc issues. A classical multi-level hierarchy is shown in Figure 1. The hierarchical levels are organized from bottom to top as a function of sensory data abstraction and from top to bottom as a function of control specificity. In general, the higher a particular level, the slower is its update rate. A natural consequence of the hierarchical organization is that in each higher level, the representation of the dynamics becomes increasingly abstract (symbolic), control actions change from simple signals to complex data structures, and the control processing becomes more complex, often declarative rather than algorithmic. The top of the hierarchy is the man-machine interface, where goals are set for the system and status is returned. The lowest levels are the interface to the real-world via sensors and actuators and represents the only way in which the system can effect the world. Today, system designers are responsible for including design features such as layer definition and functionality, data representation, computational approach, system stability, reliability, *Boeing Advanced Technology Center I World I Figure I . Hierarchical Intelligent Control System robustness and accuracy. It is our contention that a controller can be defined for each level of the hierarchy from the top level goal specification to the bottom level sensors and actuators. Given this framework, a mapping of the controller into an integrating architecture is also necessary. Issues of update rate, knowledge representation, communication, and execution rate are the driving r e q u i r e m e n t s .