Humans have a strong natural ability to remember item locations. In graphical user interfaces, this ability is one of the primary mechanisms by which users become efficient. However, there are two ways in which modern applications often fail to exploit the potential of spatial memory. First, they overuse hierarchical structures such as cascading menus, which slows down interaction for expert users who already know item locations; and second, they move items around, most commonly in response to changing display geometry. The three goals of this thesis are therefore to (1) develop a better understanding of human spatial memory in the context of user interfaces; (2) design and validate efficient command-selection interfaces based on the strength of spatial memory; and (3) design and validate interface strategies that allow users to maintain spatial memory even when display geometry changes.

Addressing goal (1), a comprehensive literature review of spatial memory for user interfaces is presented. The review covers underlying psychological models of spatial memory, the observable properties of spatial memory, and existing applications of spatial memory to human-computer interaction. In addition to informing the research in this thesis, the review is intended to provide a useful summary of the state of spatial memory research for scientists in HCI, as well as providing a set of design guidelines on spatial memory for practitioners.

Addressing goal (2), this thesis presents the design and evaluation of two related user interface techniques, CommandMaps and StencilMaps. The CommandMap is a spatially stable interface with a flattened hierarchy, intended as a replacement for cascading menu systems. Theoretical performance predictions indicate that CommandMaps should be significantly faster than traditional user interfaces such as menus and the Microsoft Office Ribbon, and laboratory-based empirical studies of command selection confirm these predictions. These positive results motivated the design and implementation of two real-world CommandMap user interfaces based on Microsoft Word and Pinta (an open-source image editing application). Evaluation results confirmed that CommandMaps continue to demonstrate performance and subjective advantages in the context of actual tasks, including interleaved command selection, typing, and direct manipulation. Qualitative data gathered from interviews, questionnaires, and conversations provide substantial insight into users' reactions to CommandMaps, leading to a set of design recommendations regarding when and how they should be implemented in real applications.

One design limitation identified during CommandMap evaluations was that novice users could be initially overwhelmed by the number of controls displayed at once. To address this concern, an extension to the CommandMap, called a StencilMap, was designed and evaluated. By using a stencil overlay to de-emphasise more advanced controls, the StencilMap directs users' visual search to a subset of controls they are most likely to need. Then, when novice users progress to the full interface, they can utilise their existing knowledge of command locations. An initial study shows that stencils are more effective at guiding visual search than ephemeral adaptation, another subset emphasis technique; however, users' spatial learning decreases as the amount of guidance increases. A second study compared StencilMaps to a palette-based subset interface, which displays the most likely commands in a ready-to-hand tool panel. Results show that StencilMaps enable stronger learning of the full UI compared to the palette approach.

Addressing goal (3), this thesis presents an investigation of how interfaces can be adapted to changing interface constraints while still supporting the user's memory for item locations. A human factors study on spatially consistent transformations was conducted, with results showing that people's spatial memory is only minimally disrupted by geometric transformations (such as scaling, translation, or perspective distortion), as long as the set of items in a display is transformed as a whole. This idea is then applied to a file browser layout: by scaling the item grid when the parent window is resized, rather than reflowing items, memory for item locations can be maintained. A second study validates this idea, showing that a scaling interface outperforms both reflow and scrolling-based techniques for revisitation when windows are resized.

In summary, the contributions of this thesis are: (1) an in-depth literature review of spatial memory in psychology and HCI, which is intended to inform designers and future researchers as well as the material in this thesis; (2) the design, implementation and evaluation of a new interface, the CommandMap, which shows that spatial stability and hierarchy flattening enable a high ceiling of expert performance; (3) the design of a stencil overlay technique to help novice users find commands, and an evaluation highlighting the key trade-off between helping users and allowing them to learn; and (4) empirical evidence showing that most types of whole-interface transformations have a small effect on spatial memory, and that correspondingly, scaling interfaces outperform reflowing interfaces under changing window constraints.