What is a cognitive map? Unravelling its mystery using robots

Despite years of research into cognitive mapping, the process remains controversial and little understood. A computational theory of cognitive mapping is needed, but developing it is difficult due to the lack of a clear interpretation of the empirical findings. For example, without knowing what a cognitive map is or how landmarks are defined, how does one develop a computational theory for it? We thus face the conundrum of trying to develop a theory without knowing what is computed. In this paper, we overcome the conundrum by abandoning the idea that the process begins by integrating successive views to form a global map of the environment experienced. Instead, we argue that cognitive mapping begins by remembering views as local maps and we empower a mobile robot with the process and study its behaviour as it acquires its "cognitive map". Our results show that what is computed initially could be described as a "route" map and from it, some form of a "survey map" can be computed. The latter, as it turns out, bears much of the characteristics of a cognitive map. Based on our findings, we discuss what a cognitive map is, how cognitive mapping evolves and why such a process also supports the perception of a stable world.