How Albot1 Computes Its Enduring Maps and Finds Its Way Home
MetadataShow full metadata
This thesis continues the experimentation with Albot1, the second in the series of Albots, in search for a theory of cognitive mapping. Albots are robots created in the Centre for Artificial Intelligence Research (CAIR) Laboratory at AUT for investigating hard problems in cognitive sciences. Using Albot1, Yeap and Hossain (2019) developed a theory of cognitive mapping that explains what kind of a global map is computed and why. However, they also leave open a puzzle: why couldn’t one use the transient egocentric global map computed to detect that one is still moving in the same local space? Failure to do so would limit the usefulness of the theory itself. The goal of this thesis is to show how the theory could be extended to overcome this puzzlement. To find a cognitively interesting solution, I continue to empower Albot1 to perform tasks that are deemed to be most relevant to cognitive mapping at this level. If a species were to compute a route map and an egocentric map as suggested in the theory, what is next? What would be computed by, say, a more advanced species and why? I began by empowering Albot1 to find its way home using the maps computed and in particular, could Albot1 discover the use of short-cuts? The first experimentation, consisting of nine experiments, leads to the use of a trace map for returning home. The latter is a global map of key points in the journey that can direct one to return home. It is generated directly from the route map and interestingly, one could also detect short-cuts using it. However, its use is limited as it is still a global map and therefore its accuracy is an important factor to get one home. The second experimentation, consisting of three experiments, focuses on studying the use of a more enduring map in cognitive mapping. What does it mean to have an enduring map? How is the process of computing such a map interacts with that of an egocentric map? What role does it play in cognitive mapping? These are the questions being investigated. The result is the realization that computing such a map is problematic and requires the use of a new set of mechanisms that are uncommon among species. It will also lead to computing a more precise and detailed global map. Empirical researchers have shown the common use of geometry of a place to re-orient in a local environment among a variety of species. This suggests that computing the geometry of a place could be the first step towards computing a more enduring representation in cognitive mapping. The third experimentation, consisting of two experiments, investigated this possibility. Could computing the geometry of a place lead to a more enduring representation that could be used to find one’s way in the environment? Could one discover short-cuts using these representations? The result leads to computing a useful place map from the route map. The former is an abstract representation that could help one to re-orient in a place and to find short-cuts home. It is not a precise map and computing it is straightforward, thereby overcoming the problems one faced earlier. A major contribution of this work is the successful extension of Yeap and Hossain’s (2019) theory, not only to overcome one of its severe limitations but also in ways that are cognitively interesting. The extension of the theory to compute either or both a trace map and/or a place map from the route map means that what is computed initially (i.e. the route map) is rich enough to support cognitive mapping. Both representations support the use of short-cuts to go home and they are computed directly from the route map using simple mechanisms that are found in many different species.