Ralph: Architecture Background

The Ralph agent starts with the idea of an ideal decision-theoretic framework, where the agent will always choose the best possible action with respect to its goals. Realizing that this ideal is not realistically possible, the Ralph architecture attempt to come as close to it as possible through the use of two specific architectural components, Multiple Execution Architectures, and meta-level control of planning.

These techniques are demonstrated using an autonomous underwater vehicle. This particular domain requires real-time, complex behavior, and the consequences of inappropriate behavior are costly. In addition, the agent must be able to act independently for extended periods of time. For these reasons this is seen as an ideal testbed for the Ralph Architecture.


Detail from Ralph