Centrality vs Modularity

Centralized architectures are ones in which one unit (or module) does most of the cognitive work of the agent. The advantages of having one module control of most functionality are:

1. interface issues disappear,
2. there isn't a need to duplicate functionality across modules,
3. there may be greater efficiency.

Modularized architectures are ones which assign different small tasks to different units (or module) in the agent. To achieve a high-level goal the architecture must break it down into parts for the appropriate modules, and must then integrate the results of the modules into the appropriate actions. If the modules are arranged in a hierarchical level of control the architecture is called layered.

The advantages of having several modules are:

1. architectures are easier to modify and maintain
2. architectures may be easier to understand

Press this line for general discussion on modular architectures.



 Examples of centralized architectures are:

SOAR by A. Newell et al.
Teton by VanLehn & Ball.
Theo by T. Mitchell et al.



 Examples of modularized architectures are:

Atlantis by E. Gat
Behavior-Based Programming by R. Brooks.
Dynamic Control Architecture by B. Hayes-Roth.
ERE by Drummond et al.
Homer by Vere & Bickmore.
Icarus by Langley.
Prodigy by Carbonell et al
RALPH by Ogasawara and Russell.
Subsumption Architecture by R. Brooks




 Examples of architectures that make no commitment are:

MAX by D. Kuokka





Other Properties.

Back to the Title Page.