In our fragmented environment, spatial information is dispersed across various locations, requiring us to rely on spatial memory to navigate and find shortcuts. This thesis explores how our spatial memory for navigable space, or survey knowledge, is structured. It contrasts two main theoretical approaches: Euclidean map approaches, which suggest spatial locations are represented in a global, map-like format, and enriched graph approaches, which propose a network of connected, partitioned places. Through four studies, participants learned spatial relations in virtual environments and completed survey tasks, such as pointing to object locations from memory. Results indicate that our memory of navigable space aligns more closely with the enriched graph model—a network of local places linked by directed connections, without needing a global reference. Survey estimates appear to be constructed incrementally based on memorized connectivity and are generally transient. Moreover, participants can acquire a general reference direction that applies to a region or the entire environment. Ultimately, our representation of navigable space resembles an impossible puzzle, where memorized pieces and connections do not align globally.
