Issues

• Does the episodic memory lead to performance degradation?

  Homer's episodic memory provides many useful capabilities. However, it must have a negative effect on scalability. As the agent gains more and more experience, it becomes increasingly difficult for the planner to find relevant events in its history. This is a classic example of the utility problem, since knowledge is merely amassed and never forgotten.

• Does modularity really promote efficiency?

  Homer's modular organization made it very easy to incorporate a natural language interpreter and generator. However, it made a uniform representation of and universal access to knowledge impossible. Presumably, this allows more specialized methods designed to promote efficiency, but this is questionable. It is also possible that interaction between modules may provide larger context information and thus guide the processing of each toward more relevant areas, improving overall efficiency.

• Do the results extend to the real world?

  Currently, all basic agent sensing and manipulation takes place on a simulator. Whenever simulators are used, the results are highly sensitive to the accuracy of the model. It is difficult to extend simulated results to behavior in the real world, so this problem will remain until Homer actually has an opportunity to operate in the ocean.