CHI 90 Pmceednrs April 1990 THE DESIGN SPACE OF INPUT DEVICES Stuart K. Card, Jock D. Muckinlay, and George G. Robertson Xerox Palo Alto Research Center 3333 Coyote Hill Road Palo Alto, CA 94304 415-494-4362, [email protected] ABSTRACT most engineering disciplines proceed to organize these A bewildering variety of devices for communication designs by developing abstractions, theories, or other from humans to computers now exists on the market. organizing principles in order to get a view of the design In order to make sense of this variety, and to aid in the space behind these and potentially other designs (e.g., design of new input devices, we propose a framework [21]). Previous work on human-machine input devices for describing and analyzing input devices. Following has provided three lines of development in this area: Mackinlay’s semantic analysis of the design space for toolkits, taxonomies, and performance studies. graphical presentations, our goal is to provide tools for the generation and test of input device designs. The de- Tool&i. User interface toolkits or user interface man- scriptive tools we have created allow us to describe the agement systems help with a wide range of problems in- semantics of a device and measure its expressiveness. cluding the construction, runtime execution, and post- Using these tools, we have built a taxonomy of input runtime analysis of a user interface [22]. They may help devices that goes beyond earlier taxonomies of Buxton systematize input device knowledge by providing a li- & Baecker and Foley, Wallace, & Chan. In this paper, brary of pre-built input device modules [18, 161, archi- we build on these descriptive tools, and proceed to the tecture and specification techniques for combining these use of human performance theories and data for evalu- modules [2, 231, or post-processing analysis tools 1171. ation of the efleciiveness of points in this design space. Sometimes, as in Anson [2], they even provide archi- We focus on two figures of merit, footprint and band- tectural models of input device interactions. But the width, to illustrate this evaluation. The result is the device models implicit in user interface toolkits sketch systematic integration of methods for both generating only a limited picture of the design space of input de- and testing the design space of input devices. vices and their properties. Even for the construction of interfaces, they present interface designers with many KEYWORDS: Input devices, semantics, design knowl- design alternatives, but do little to help with the de- edge systematization. sign decisions themselves. In order to achieve a sys- tematic framework for input devices, toolkits need to INTRODUCTION be supported by technical abstractions about the user, Human-machine interface technology has developed to the devices themselves, and the task they are used in the point where it is appropriate to systematize existing performing. research results and craft into a body of engineering and design knowledge. A case in point is the design of input Taxonomies. Two recent attempts have been made to devices. A bewildering variety of such devices now exist provide abstractions that systematize the design space on the market, including typewriter keyboards, mice, of input devices. Foley, Wallace, and Chan [lo] focused headmice, pen and tablets, dialboxes, Polhemus cubes, on computer graphics subtasks. They classified input gloves, and body suits. Given an abundance of designs, devices under the graphics subtasks they were capable of performing (e.g., the tablet and the light pen are capable of character recognition). They also reviewed experimental evaluations of input devices. Buxton and Baecker [4, 31 have proposed a taxonomy of input de- vices classified according to the physical properties and the number of spatial dimensions they sense. The lim- Permission to copy without fee all or part of this material is granted itation of the Foley, Wallace, and Chan scheme is that provided that the copies are not made or distributed for direct the categories, while reasonable, are somewhat ad hoc commercial advantage, the ACM copyright notice and the title of and there is no attempt at defining a notion of complete- the publication and its date appear, and notice is given that copying ness for the design space. The limitation of the Buxton is by permission of the Association for Computing Machinery. To copy otherwise, or to republish requires a fee and/or specific permission. 0 1990 ACM 0-89791-345-O/90/0004-0117 1.50 cH190-ngs prxil1990 and Baecker scheme is that it only includes continuous of human performance theories and data for the evalu- devices. ation of points in this design space. The result is the systematic integration of methods for both generating Performance studies. Seve:ral studies have been made of and testing the design space. the performance of different pointing devices. English & Englebart [7] studied several devices and found the GENERATING THE DESIGN SPACE mouse the fastest device. Card, English, & Burr [5] con- Conceptually the most general case of human-machine firmed these empirical results and discovered that point- interaction is the case of a human interacting with an ing speed with the mouse is governed by Fitts’s Law [9] embedded computer (e.g., the autopilot on an airplane). with a bandwidth similar to that of the hand. Sub- Such an interaction can be modeled as the interaction sequent studies have empirically compared speed and in an artificial language among at least three agents [6]: preference of various devices ([lo, 1, 8, 11, 251. Unfor- tunately these have not always agreed and most studies have not attempted to disentangle task, subject, and 1. a human, human performance variables. 2. a user dialogue machine, and ANALYTIC DESIGN FRAMEWORK 3. an application. In order to understand how the above results could be accommodated in a single systematic framework, it is useful to consider the role of input devices in human- We can trace the semantics of an input device by trac- machine communication. An input device is part of ing the mappings from human action through mappings the means used to engage in dialogue with a computer inherent in the device and finally into changes in the pi+ or other machine. The dialogue is not, of course, in rameters of the application. natural language, but is conducted in ways peculiarly suited to interaction between human and machine. Un- There are two key ideas in modeling the language of like human-human conversation, the dialogue is between input device interaction: fundamentally dissimilar agents-both in terms of per- ception and processing. Furthermore it takes place un- 1. A primitive movement vocabulary, and der conditions (e.g., the persistance of displays) that are different from the evanescent, sequential oral converslt 2. A set of composition operators. tion that is often taken as the model for communica- tion. Instead of words, the user may move the mouse The movement vocabulary gives the elementary sen- and press buttons. Instead of words, the machine may tences that can be expressed in the artificial language. show highlighted animated diagrams. The composition operators give methods of combining this vocabulary into a combinatorically richer set. The design of human-machine dialogues is, at least in part, the design of artificial languages for this commu- Primitive Movement Vocabulary nication. Mackinlay [13, 141, in work on the automatic We begin with the observation inspired by Baecker and generation of displays, suggested that each display could Buxton [3] that: be thought of as a sentence in such a language and that such sentences could be analyzed as to their ability to transmit an intended semantic meaning from the ma- Basically, an input device is a transducer from chine to the user. This analysis has direct consequences the physical properties of the world into logical for the design of machines. Semantic theories provide values of an application. the means by which the design space can be gener- ated. Human performance studies provide the means by which design points in the space can be tested. We Formally, we represent the input device as a six-tuple can use this basic approach as a means for systematizing knowledge about human interface technology, including the integration of theoretical, human performance, and (M, In, S, R, Out, W) artifact design efforts. In particular, we can use this approach to integrate knowledge gained from toolkit, where taxonomy, and human performance literature. In an earber paper [15], we addressed the semantic anal- l M is a manipulation operator, ysis of input devices and used this to generate the design space of input devices. In this paper, we build on the l In is the input domain, results from the earlier paper and proceed to the use l S is the current state of the device, 118 Cl-II 90 Ploceedngs Ppil1990 ~~ Fipun I. Physical proper& scmed by input devices. l R is a resolution function that maps from the input domain set to the output domain set, Input l Out is the output domain set, and state Resolution fn e W is a general purpose set of device properties that OutQlJt describe additional aspects of how a device works (perhaps using production systems). Figure 1 lists the various manipulation operators pos- sible for an input device. They are an extension of the physical properties suggested by Baecker and Buxton 131. They represent all combinations of linear and ro- Merge composition is the combining of two devices such tatary, absolute and relative, position and force. Al- that the resulting input domain set is the cross product though other input devices are possible (based, say on of the input domains of the two devices. A mouse, for speech or heat), virtually all input devices use some example, can be thought of as the merge composition combination of these properties.