Feedback

It would be of considerable interest to get better examples of how EM can be applied to developing strategies in games. So far, there has been little convincing application of the concepts of observation, dependency and agency in this arena. There is some kind of dependency between what the player believes about the location of the Wumpus etc and what they have observed in their previous traversals of the board. Its interesting how this kind of depencency in turn relies on knowledge about the Wumpus's own capabilities (whereas pits don't move etc). If you are really focusing on comparing strategies, then be sure that you take care to be explicit about why (if at all) EM is relevant (since this is the overall subject of the EM model!). Beware of placing your focus too narrowly on the AI problem without thinking about how your use of EM might compare with a traditional AI analysis (and treatment of knowledge in particular). It might be of interest to show how you could develop a customised notation to talk about the Wumpus game also (e.g. to express the knowledge you have informally, and pose/answer questions about the current situation).

For what it's worth, there's a very simple model of Hunt the Wumpus in the EM archive. This could be of interest if you wished to incorporate any discussion of how principles of EM apply to the more basic elements of the model-building etc. There's also an (arguably!) elegant arca model of the dodecahedron (if I remember correctly) in the directory wmb/public/projects/notations/arca/DEMOS. This model would enable you to refer directly to the location you would get to by following a specified sequence of edges from the current location.

More scholarly AI references would be useful.