This thesis describes the implementation of Albot1, the second in the series of Albots. Albots are robots created in Centre for Artificial Intelligence Research (CAIR) Laboratory at AUT for investigating hard problems in cognitive science concerning spatial cognition. To create an Albot, a computational theory is first proposed and then implemented on a mobile robot. Its behavior is then studied to gain further insights into the underlying process. With Albot1, the problem being studied is: how an imprecise and incomplete map is computed without integrating every successive view of the environment. Albot1 is based on Yeap's (Yeap2011a) computational theory of perceptual mapping.

That this problem is both worth studying and hard is evident in the controversial debate among psychologists as to the nature of such a map. Despite much interest, the notion of a cognitive map remains vague and controversial. Furthermore, robotics research has shown that the traditional process of integrating every successive view to form a map is highly susceptible to cumulative errors coming from the sensors. These errors must be corrected in order to produce a useful map, but in doing so, what is produced is a precise and complete map. Humans don't have such a map in their head but then, how and what kind of map is produced?

Yeap's (2011a) theory suggests that a map is formed from integrating views representing local environments visited. In his model, viewers don't track their position in the map, nor are errors from the sensors corrected. Rather, they attend to landmarks and remember the local environment that they are about to explore. Describing these local environments using a single global frame of reference gives one a map of the entire environment traversed.

The theory was tested with a robot equipped with a laser and an odometer. Six experiments were conducted to evaluate both the process and the map computed. Specifically, the experiments evaluated: (i) Albot1's ability to compute a map in an office-like environment, (ii) Albot1's ability to orient itself, (iii) Albot1's ability to handle errors, (iv) Albot1's ability to compute its map using continuous motion, (v), and (vi) Albot1's performance compared with SLAM-based algorithms.

The results provide confirmatory evidence that the process is both robust and useful. It successfully captures the overall shape of the environment traversed as long as no parts are being re-visited. Once a familiar part is being re-visited, the previous experience of that part of the environment is wiped from the map and overridden. The map computed is thus both a transient and a dynamic map. It represents a trace of one's experience through the environment rather than a complete map of the environment experienced. Having successfully computed a transient map, I also investigated how a more enduring map, in the form of a topological network of places, could be built. A further three experiments were conducted to evaluate Albot1's "cognitive map".

The successful implementation of Albot1 introduces a new paradigm for robot mapping and provides significant insights into how Yeap's (2011a) perceptual mapping works.