The RALPH-MEA agent was tested in a simulated but dynamic underwater environment. The RALPH architecture makes certain strong assumptions about the world that are most easily satisfied in the constrained environment of a simulator. The decision-theoretic execution architecture requires that a utility function be available which can determine the utility of every possible state. Obviously, the easiest way to implement this is to know every possible state in advance. Another strong assumption is the Markov Assumption, which makes each world state dependent only on the state that immediately preceded it.
Go to a discussion of this environment for multiple architectures.