RALPH - Multiple Knowledge Types

Description

• RALPH utilizes different knowledge types in each execution architecture in order to produce decisions that can differ in the tradeoff between quality and speed. Knowledge types are labelled by letters and the current time is labelled as "t":
  • A: Information about the current state at the current time t, including sensor-based information
  • B: Knowledge about the effects of actions
  • C: The utility of a projected state (time t+1) based on knowledge of actions and the environment
  • D: Best action choice directly from current state. This knowledge is similar to condition-action rules used in production systems. It requires very little reasoning, and is thus utilized when reactivity is necessary.
  • E: The utility of the state at time t+1 acquired without explicitly reasoning about the composition of the future state.
  • F: Best action choice directly from projected state.