The tests demonstrate that an effective and accurate memory model can be developed for inaccessible environments. The model uses a probabilistic approach and crucial assumptions to provide a predictive aspect to model the unseen parts of the state. The model considerably outperformed a simpler approach with only a small loss in accuracy. As mentioned before, even though this model was designed and tested using the older version of the simulator, equations are provided in the appendix that use the added information of the simulator. These equations, coupled with the descriptions throughout this paper, should allow other researchers to implement our memory model for use in the Soccer Server system.
Further research in this area could look at applying learning techniques to the predictive aspects of the model. The client could learn the effects of actions and apply it to situations where there is no information for prediction (i.e. situations when there are no markers visible.) Other research could focus on designing behaviors for the clients, which take advantage of the model.