Motor Skill Learning in User Interfaces via Discretized Pie Menus
An important area of human-computer interaction research focuses on pointer-based graphical user interfaces. The literature shows that designers must be sensitive to the subtleties resulting from the interface paradigm. Specifically, video game designers must understand the level of difficulty and thus the overall efficiency of a user interface. Can the player, in quick succession, efficiently perform many distinct actions using a traditional pointer-based GUI? Are there too many icons to place on one screen? What if the player uses a gamepad instead of a mouse?
This article provides a short overview of Fitts' Law and related material as a basis for discerning common pitfalls of pointer-based interfaces. Discretized pie menus are introduced as an alternative to the pointer interface. Following that is a discussion of its merits and demerits.
Although this article uses typical real time strategy games as an example, the ideas presented can be generalized to any traditional pointer-based game interface.
Common Pitfalls of Pointer-Based Interfaces
Fitts' Law is a logarithmic function describing the index of difficulty (ID) in movement based motor tasks [4, 5]. The function is defined as
where A is the amplitude or distance of movement and W is the width of the target where movement ends. In Figure 1, if a subject were to perform a one dimensional, horizontal movement from the start position (marked S), to any point within the rectangular region, then A is the distance to the rectangular region and W is the width of the rectangular region.
Many HCI researchers have used Fitts' Law  with one of the first uses emerging from the work of Card, English, and Burr . With the aid of Fitts' Law, Card et al. evaluated the performance of four devices in text selection on a CRT screen. Of the four devices (mouse, rate-controlled isometric joystick, step keys, text keys) they found the mouse to provide the best performance. Card et al. used Welford's formulation of Fitts' Law , which expressed the time required to make a hand movement. Given an index of difficulty (ID), it is possible to predict the movement time based on the following equation :
where a and b are constants depending on the type of device used (e.g., how long it takes to grasp the device).
Though Fitts' Law deals with movements in one dimension, it has been shown by Mackenzie  that the law can be extended to two dimensions. Mackenzie suggests that a more accurate strategy for determining W (target width) in two dimensions is to use either a) the "Smaller-Of" model where W is the smaller of the width or height or b) the " W' " model where the 1D nature of Fitts' Law is maintained by measuring W along the approach (to target) vector. Clearly, Fitts' Law is applicable to two dimensional user interfaces.
Note that the major factors in determining the difficulty of movement based actions is the amplitude (distance to target) and the width (size of target). These spatial factors contribute to subtle deficiencies in typical pointer-based interfaces.
Real time strategy games commonly implement an interface where players use a pointer to select icons representing actions or units. The result is an easy to learn interface requiring little explanation. However, the downside of such an interface emerges when players are required to perform several actions in quick succession. For example, if the player wishes to quickly build several turrets, he/she would repeatedly reposition the pointer from the game board to the menu icons and back. From the experiments performed by Card et al. , mouse movements over short distances (1-16 cm) required just over one second of positioning time! Given that data, a player wishing to build in quick succession would require a considerable amount of time.
Designers solved this problem by providing shortcut keys to frequently used actions or units. This alleviates the time cost associated with pointer based interfaces. Given Fitts' Law, A = 0 and so MT = a. The time to perform the action is simply the time it takes for the player to press a key.
Yet in resolving the time issue, new problems arise. Now a player must follow an unrelated input protocol requiring the memorization of shortcut keys. Over time, veteran players run the risk of forgetting these keys. Worse, this solution is not suitable to game platforms which do not include a keyboard (consoles and portable gaming systems). It also does not scale to games requiring many actions and/or units as it would either clutter the screen with icons or require additional pointer movements (i.e., submenus). One way to alleviate these shortcomings is to teach an interface relying on motor skills.
Motor Skill Learning in User Interfaces
In studying H. M., a 40-year-old patient showing a severe anterograde amnesia, Corkin  discovered that despite the patient's inability to remember new information, he was able to acquire new motor skills through repeated practice. His findings supported the theory that a motor skill memory system exists separate from other memory systems. Corkin  also states:
In general, retention of a variety of motor skills is very high, even for no-practice intervals of up to two years, and relearning is rapid (for reviews see , Chap. 8, and ). On the other hand, verbal material is typically forgotten much more quickly [8, 10]. The few instances in which motor learning dissipates rapidly (, pp. 267-237; ) are those in which the task has a large verbal or other non-motor component.
The fallibility of human memory can be offset by taking advantage of the robustness in motor skill learning.
It can be argued that shortcut keys rely on motor skill learning. Since touch-typing is a motor skill, why do shortcut keys not exhibit the robustness described by Corkin? The problem emerges when the act of typing is contrasted with the act of pressing a shortcut key in a typical RTS game. With typing, users always place their fingers on the "home" keys thereby allowing them to perform each action in a consistent matter. In other words, they learn to type each letter by repeatedly practicing the same motion. With a typical RTS game, the player may approach the key from various angles and positions. Unlike typing, there is no obvious "home" position that allows the player to repeatedly practice the same motion. Thus the player cannot rely solely on his/her motor skill and is forced to use a less robust memory system. Providing a "home" position for the user enhances the motor skill element of the interface learning.
The directional/movement keys are one of the most familiar "home" positions in gaming. Typically (assuming, without loss of generality, the player is right handed), the middle finger is on the up-arrow key with the index and ring finger on the left and right arrow keys respectively. This position can be transposed to other areas of the keyboard (e.g., the WASD keys). On a gamepad, the natural position is the thumb on the directional-pad (D-Pad). Taking advantage of these natural positions activates players' robust motor skill learning mechanisms in the brain.
Motor Skill Learning in Discretized Pie Menus
Many games, most notably "The Sims", now employ a pie menu user interface. It is easy for novice users, who simply follow the menu labels, and remains efficient for expert users, who remember the correct sequence of movements without reading the labels . In addition, novice users do not need to learn a separate input protocol in becoming an expert user. The graphical display is also kept clean since the menu is drawn only when the player performs a menu action. For games requiring a vast number of options, multi-level or hierarchical menus can be used, an example of which can be seen in the Half-life mod, "Natural Selection".
Still, pie menus rely on pointer-based input devices and thus distance and target size become factors of MT. Using directional keys to discretize the menu interface gains the benefit of minimizing MT (do not need to take into account distance and target size) while continuing to leverage the robust nature of motor skill learning. In addition, the interface can be utilized on any gaming platforms. As an example, suppose a player wishes to place a missile turret in a typical RTS game using a gamepad and discretized pie menus. The player might take the following steps (see also Figure 2):
The player is able to perform the actions required in RTS games without the aid of a mouse.
Caveats of Discretized Pie Menus
In practice, there are various caveats to the proposed interface system. Depending on the game, the number of key presses required to navigate a multilevel menu may increase MT to an unacceptable amount. This problem can be circumvented by providing shortcut keys to sublevels within the menu. Assigning the shortcut keys such that it suggests an obvious "home" position on the keyboard/gamepad enable the user to continue motor skill learning. Obviously, forcing the user to learn too many shortcut keys will attenuate the effectiveness of the interface.
Anecdotal evidence also suggest discretized pie menus have a short but very steep learning curve. This is possibly because the interface requires an input protocol that vastly differs from other traditional input protocols. Therefore the difficulty may drop as more developers adopt this scheme. Ultimately however, the difficulty depends on the difficulty of the game. Since all actions required in a game can be placed in one hierarchical pie menu, it easy to overwhelm the novice player with too many options.
Despite these caveats, discretized pie menus keep the game interface clean while allowing players to learn using robust motor skill memory. Once players overcome a short learning curve, they can perform actions quickly and efficiently. Finally, discretized pie menus are scalable to different input devices including mouse, keyboard, and gamepad.
About the Author
Jonathan Mak is an independent developer exploring phenomena in player-game interactions. He is currently developing the abstract, action-strategy game, Gate 88.