Jump to the ERE Overview page.
Jump to the Architectural Details Menu page.
Jump to the Home page.

Architectural Details

The ERE architecture uses various problem-solving methods such as: problem-reduction, temporal projection, and rule-based execution. These methods are implemented in three major architectural components:

Architectural Schematic

No sketch of the architecture was provided in the paper, but the following diagram represents my understanding of its structure. It depicts the dependencies of the modules on each other, the knowledge dependancies of each module, the control dependencies of the each modules (see Taskability in the Capabilities menu for a discussion on this), and the interaction of the modules.

Knowledge Types

The user must provide knowledge to the system. This knowledge is used by several, but not all, of these architectural components. There are four types of knowledge, which are as follows:

Shared Knowledge

As mentioned, this knowledge is shared between the components of the system: strategies affect the Reductor and the Projector; SCRs affect both the Projector and the Reactor. Compiled knowledge (SCRs) is also shared. Furthermore all of these are ultimately dependant on the Causal Theory. All of this brings up an important: if knowledge is changed via user input or learning, then there can be reprecussions throughout the system. Knowledge dependencies should be maintained to addressed the problem. This is an area of continued research in ERE.

Anytime Operation

The reasoning mechanisms found in ERE provide capabilities not found in other systems. ERE can produce plan in an anytime fashion: the Reductor and Projector modules produce synthesized advice (strategies for the Projector; SCRs for the Reactor) for their subsequent module within the time constraints of the system. ERE is the first architecture to produce such "reaction plans" that provide an appropriate balance between limited and absolute robustness.