Design Integration Laboratory

Three-Dimensional Sketching

Architectural Design Software Proposal - 1988

Note: This is a rough web posting from an automatic conversion program, containing formating, reference location, and typographic errors from the translation process, and missing some illustrations.

Three-Dimensional Sketching

Master's Project
Master of Architecture Degree Program
Option III

College of Environmental Design
Department of Architecture
University of California, Berkeley

Kevin Matthews

August, 1988

Mark Smith, Assistant Professor of Architecture
Fred Stitt, Visiting Lecturer in Architecture

Three-Dimensional Sketching


Abstract .. 1

Introduction .. 2

The Problem

Design with Computers? .. 2

The Context

Getting ideas down without getting distracted .. 7

Design and Computing

An Interface Issue .. 15

Existing Two-Dimensional Interfaces .. 17

Existing Three-Dimensional Interfaces .. 22

Three-Dimensional Sketching .. 27

Putting It All Together

Looking Forward .. 40

Conclusion .. 42

References .. 44

© 1988 by Kevin Matthews. All rights reserved

Master's Project August 1988
Department of Architecture
University of California, Berkeley
Kevin Matthews

Three-Dimensional Sketching


Graphically oriented computers have become an efficient means for the production of construction documents, but despite recent large advances in computer graphics capabilities computers are still not effective tools for creative design work. This paper will examine the design process to determine what has made it resistant to computer assistance, and then discuss how to address the unique nature of architectural design work by changing the character of human-computer interaction. Observations about existing design process based on sketching with paper and pencil will show that the design utility of computer modeling systems can be advanced through appropriate innovations in interface design.

The use of sketch modeling in early design is a project distinct from the use of more elaborate computer models as visualization tools during design development. While imaging techniques have advanced, the methods for creating the fundamental three-dimensional model have lagged. The model shop metaphor suggested by previous research on architectural design interfaces is not appropriate to the crucial design phase of actually generating and capturing new ideas. In conventional practice this work is not done in a model shop, where tooling operations must be done in a premeditated sequence. Rather, early design is done freehand on paper. In general, we must look to the freehand sketching process for the best clues to a successful design interface.

The paper will examine the sketching process and delineate its fundamental characteristics. These provide the basis for a new approach to a three dimensional freehand sketching environment optimized for architectural design. The modeling space is characterized by a noun-verb syntax direct -manipulation interface, using direct mouse control of a crosshair in three dimensions and objects with click-selectable handles to achieve facile editing in three dimensions. In addition, some computer modeling tools will be suggested which would facilitate the seamless refinement of conceptual sketch models into precision models for architectural design development and design documentation.


Computers are now beneficial for the production of construction documents, but not usable for architectural design despite recent large advances in computer graphics capabilities. This paper will examine the creative design process to determine what has made it resistant to computer assistance, and then discuss how to address the unique nature of architectural design work in the character of human-computer interaction. A few key observations about existing design process will lead to the argument that a major advance in the design utility of computer modeling systems is accessible through appropriate innovations in interface design.


The Problem

Design with Computers?

Graphic computing has tremendous potential for aiding predictive visualization, which is central to the creative process of building. With the recent advent of fast raster graphics on expansive high-resolution color monitors, powered by quick processors and growing amounts of random access memory, improvements in computer imaging have spawned wide ranging new uses. These have come to public attention in computer animations on television and through the commercial success of the graphically oriented Apple Macintosh microcomputers. It is now possible, with a computer graphics system affordable to a medium sized architecture firm, to create a three-dimensional data model of a building--including landforms, vegetation, fenestration, exterior detailing, and interiors with furnishings--that can be viewed from any distance or direction, and automatically rendered in a few minutes in full shaded color with accurately computed shadow casting (figure 1).

Figure 1. Color rendering from a GDS three-dimensional model.

Leading firms have even used special hardware and software to produce full perspective interior and exterior "fly throughs" for video presentations. [HOK Computer Systems: demonstration video tape.]

These powerful tools for looking at buildings before they are built transcend traditional presentation techniques such as physical models and hand renderings because the computer's ability to rapidly re-image a model space from any position virtually eliminates viewing restrictions due to scale or position. Unlike most existing forms of manual presentation, a computer model may be adjusted interactively. This opens new avenues for the exploration of a building while it is still "on paper," and for as yet unexplored modes of architect-client exchange.

The brave new world is still far from unlimited, however. While imaging techniques have advanced, the methods for creating the fundamental three-dimensional model have lagged. For instance, rather than draw a human figure directly into a database, commercial animators first model it physically in foam, then trace cross sections of the foam with manual digitizers. [Computer Magic, video tape.] The duplication of effort involved in essentially modeling the figure twice is certainly inefficient, but it exemplifies the difficulty of making a computer model of a complex object.

In contrast to modeling a human figure for animation, design specifically calls for the modeling of as-yet-unknown objects. Design is more or less the creation and delineation of the model from scratch. Because design work starts with a mental construct rather than a physical one, the nature of the initial modeling process is far more critical to successful computer-aided design than is the quality of subsequent sophisticated imaging. The most spectacular successes of computer graphics have been of limited value for conceptual design work. Even in an advanced heavily computerized firm, design concepts will be developed from sketches on paper into simple massing models or drawings before a computer model is somewhat laboriously created for use in refining a project during design development. Yet it is in the earliest stages of design, when fundamental forms are being established, that the ability to think and see in three dimensions is most important. [Schilling, Terrance G. and Patricia M. Schilling. Intelligent Drawings. New York: McGraw Hill, 1987, 17, 236.]

[Williams, Greg. "The Macintosh Computer." BYTE, February 1984, 30-54.] Two-dimensional drawing with computers has advanced recently, with the windows, icons, mouse and pointers interface of the Macintosh microcomputer, manifesting research done at SRI and Xerox PARC, leading the way. [Editors of Time-Life Books. Input/Output. New York: Time-Life, 1986, 61-75.] There is still a division in ease of use between two-dimensional applications used for general drawing tasks (i.e., MacDraw) and those used for

Figure 2. Railroad diagram of a complex GDS command.

precise drafting work ("CAD applications"). But while interfaces for two-dimensional graphics have advanced dramatically, three-dimensional computer model-making remains an exclusive bastion of complex transformations and typed numerical data entry. The syntactic complexity, alone, of most existing modeling interfaces would make them nearly useless for the most creative phases of design (figure 2). Because of the awkwardness of three-dimensional modeling, paper and pencil remain supreme for early conceptualization. [Stitt, Fred A. Systems Drafting. New York: McGraw Hill, 1980, 231.]

One of the ramifications of the continuing reign of paper-based sketching for design is that for any project to be developed or produced with CAD it must go through an awkward transition, in which work on paper is duplicated by some sort of computer data entry process. [Schilling, Terrance G. and Patricia M. Schilling. Intelligent Drawings. New York: McGraw Hill, 1987, 186.] It would be much more efficient to be able to develop a project seamlessly, without the redundant work attendant upon switching media. The time and attention saved could possibly be spent on extended problem-solving.

But if computers are not yet usable for design, why not? To answer this we need to know how architects design. It seems that the traditional sketch is highly suited to the designer's needs. An examination of what is so apt about sketching will supply us with some fundamental criteria for designing a computer-based modeler that really would work for design.

I believe it is possible to create a modeling system with an easily manipulable plastic modeling environment, on existing hardware, that can be easily integrated into the already evolving, more precise and rigid production-oriented CAD environment. This will allow new approaches to several important bottlenecks in the making of buildings. Such a system will combine the ease of use and communicative density of traditional sketching with the malleable multiple dimensions of a mathematical design space. I propose this solid sketching as an achievable computer graphics application type for architectural design.


The Context

Getting ideas down without getting distracted

The schematic of a design oriented modeling system must be based first and foremost on an understanding of how people design. In general, architects design by sketching. The pencil on paper sketch provides a highly developed method of representing ambiguous ideas quickly and easily, without distracting the mind from the intuitive state required for creation and visualization (figure 3). [Kostoff, Spiro. A History of Architecture. New York: Oxford University Press, 1985, 4-5.] A useful solid sketching system will intensify the fundamental intuitive relationship between the designer and his sketch pad, rather than divert the designer from his most subtle mental state.

This intuitive state is the subject of quite a bit of philosophical controversy. [Heidegger, Martin (translated by John Macquarrie and Edward Robinson). Being and Time. New York: Harper and Row, 1962, 187.] Nonetheless, it must be examined, and conclusions about it must be drawn, in order to proceed with understanding the mechanisms by which the sketch facilitates design. These are the mechanisms from which, by a sort of reverse engineering, we will derive generalized requirements for the modeling interface.

Good design is different from simple rational thought, at least as rational thought is generally envisioned. Where simple rational thinking proceeds by handling one idea at a time it a mannerly, logical way, design must handle an overwhelming and fundame ntally inseparable welter of ideas, images, and data of non-comparable sorts. Design is concerned with what Horst Rittel has called "wicked problems." [Rittel, Horst W. J. "Some Principles for The Design of an Educational System for Design." DMG Newsletter, Vol. 4, Number 12, December 1970.]

Figure 3. The rough lines of this Aalto sketch convey a tremendous amount of ambiguous information. [Fleig, Karl, Editor. Alvar Aalto. Zurich: Verlag, 1971, 188.]
Given the multi-level ambiguity of design work itself, it is not surprising that design education has been recognized as a difficult matter, constituted of, as in architect Joseph Esherick's paraphase of noted educator Donald Schon, "mucking about in the swamp." [Schon, Donald A. "The Architectural Studio as an Exemplar of Education for Reflection in Action." Journal of Architectural Education, Fall 1984.] The swamp referred to is the "low ground" of real design problems, where our feet tend to get stuck in the rich mud of assorted decomposing but utterly vital concepts.

Yet people are particularly well equipped to deal with just the sorts of situations, complicated and without closure, which characterize the real world (as opposed for instance to the suspiciously clear and simple world of the digital computer). In fact, the real world of architectural design problems is so far beyond simple rationality that unutterability is a central characteristic of great solutions. It is the work of artists, with architects numbered among them, to succeed at handling just such intractable problems. [Alexander, Christopher. The Nature of Order. Berkeley, CA: manuscript, 1986.] [Bonta, Juan Pablo. Architecture and Its Interpretation. New York: Rizzoli, 1979.]

Moreover, since architectural meaning is embodied in the synthetic interweaving of functional structures with ornamentation, architectural design is concerned with an incredible array of parameters. The combination of doing a job while doing art, or sometimes even doing a job by doing art, makes architectural design problems even more complex, more simultaneously constrained and open ended, than those usually faced (for instance) either by artists or by engineers.

There is a particular kind of mental state, or a particular way of thinking about deep problems, that gets the normal verbal mind out of the way, and lets us work on issues en masse, by means of applied intuition. This kind of thinking works much faster than words can be formulated. At shallower levels it can be felt as visual thinking. At deeper levels, it can be sensed as the currents beneath the images of visual thinking. This type of thinking is related to the subconscious and unconscious, and unexamined, leads to many distortions in the works of confused designers.

Many great designers, including scientists, engineers, and artists, as well as architects, have discussed the importance of this intuitive thinking in their work process. For instance, Alvar Aalto spoke of the crucial intuitive aspect of his design process:

When I personally have some architectural problem to solve, I am constantly . . . faced with an obstacle difficult to surmount, a kind of "three in the morning feeling." The reason seems to be the complicated, heavy burden represented by the fact that architectural planning operates with innumerable elements which often conflict. Social, human, economic and technical demands combined with psychological questions affecting both the individual and the group, together with movements of human masses and individuals, and internal frictions all these form a complex tangle which cannot be unravelled in a rational or mechanical way. The immense number of different demands and component problems constitute a barrier from behind which it is difficult for the basic idea to emerge . . . I forget the entire mass of problems for a while, after the atmosphere of the job and the innumerable difficult requirements have sunk into my subconscious. Then I move on to a method of working which is very much like abstract art. I just draw by instinct, not architectural synthesis, but what are sometimes childlike compositions, and in this way, on this abstract basis, the main idea gradually takes shape, a kind of universal substance which helps me to bring innumerable contradictory component problems into harmony.

Malcom Quantril. Alvar Aalto: A Critical Study. New York: Shocken Books, 1983, 5.

The power of sketching for designers lies in its ability to support thinking with "negative capability," an odd but valuable term defined by the English poet John Keats in 1817:

. . . several things dovetailed in my mind, and at once it struck me, what quality went to form a Man of Achievement especially in literature and which Shakespeare possessed so enormously--I mean negative capability, that is when man is capable of being in uncertainties, mysteries, doubts, without any irritable reaching after fact and reason.

Perkins, David (ed.). English Romantic Writers. New York: Harcourt Brace Javanovich, Inc., 1967, 1209.

If we accept that much of the meaning of art affects us subconsciously, and realize that any verbal discussion of subconscious matters will never be able to achieve a completely satisfying closure, we see that extra-verbal converse is not only real and important, but actually at the heart of sensitive design.

Awareness of subconscious resonances between diverse aspects of a problem set guides the designer toward the transcendent solution. We can't work these subconscious resonances readily with usual verbal/logical consciousness--they show up in our awareness of feeling, emotion, attitude, association, and intuition. A key point, however, is that these gray areas of thought are something we can definitely tune into and use--they just require a different approach than words and numbers. [Einstein, Albert. "Autobiographical Notes," in Schilpp, P. A. (Ed.) Albert Einstein: Philosopher Scientist. Evanston, Ill: Library of Living Philosophers, 1947, 7.]

In the design of a very simple project, the extra-verbal complexity of a good solution is relatively easy to handle, because the entire solution can be conceived as a simultaneous whole. This is analogous to totally experiencing a single tree. The design can be recorded all at once as a nearly complete thing, sketched with just a few evocative strokes.

In contrast, wielding the solution images of a larger, more complex project is analogous to experiencing all the trees in a forest, as individuals, small groups, regions, and as a whole. The overwhelming size of the problem requires a system for recording fragmentary solutions as they develop, and then for massaging the fragments into a coherent whole. [Alexander, Christopher. Notes on the Synthesis of Form. Cambridge, MA: Harvard University Press, 1964.]

The two-dimensional pencil and paper sketch plays a crucial role in the early stages of a typical design process. Two-dimensional sketching uses the expressiveness of pen or pencil on paper to record ideas quickly and richly. However, paper-based sketching is limited in its facility for representing three-dimensional images, and architecture is fundamentally a three dimensional art. Furthermore, the art of drawing may become an end in itself, to the detriment of the architecture being drawn.

The sketch is nonetheless well adapted to the needs of the design process, which requires the recording of ideas and images that are ambiguous or incomplete or provisional or even too complete and vivid. Getting too specific too early freezes the creative process, so needless or excessive specificity has to be avoided. A measure of ambiguity under the control of the designer allows thoughts essential plasticity. The complex graphic shorthand of the design sketch allows the designer to put down just enough.

In addition to two-dimensional sketching, some architects use a material form of three-dimensional sketching in the early stages of design, to assist real visualization (figure 4). This

Figure 4. A physical sketch model.

Figure 5. Design development study model.

sketch modeling, using clay or foam or roughly shaped cardboard, is a special type of modeling which needs to be distinguished even from study models used during "design development" (figure 5).

In clay-based three-dimensional sketching, the necessary plasticity of the mental model is retained by the analogous malleability of the material, as well as by the vagueness of simple lumpy forms. In cardboard sketch models a provisional character is allowed by a different mechanism. We can observe a roughness of forms in torn cardboard analogous to and reminiscent of the sketchy, searching, overlapping lines of a paper design drawing.

However, sketching of early design ideas in three dimensions is limited by the actual material problems of making even the crudest physical model. For instance, if the key early image is a balcony, the physical modeler must hold the balcony image in stasis while making a physical armature to suspend the physical model balcony against gravity. In comparison, the paper sketcher can proceed in the mind's non-linear way, drawing the balcony in space with a few quick strokes, then an important tree to the side, and perhaps then the first stroke of wall.

Christopher Alexander has for several years championed another approach to design visualization. In addition to making paper sketches and physical sketch models, he and his followers have developed a complete approach to construction in which full-size mockups are made in location, to help the designers concentrate most directly on the actual reality of an actual situation. [Alexander, Christopher. The Nature of Order. Berkeley, CA: manuscript, 1986.] While this method is largely incompatible with business-as-usual in the building industry, it serves as an example of how to minimize errors of design due to faulty abstractions and misleading imagination. A full-size mockup, in place, allows the full power of applied intuition to be focused on whether the proposed solution is correct. Our modeling interface should similarly allow intuition to be applied to developing solutions.

Alexander's mockups are a sort of full-sized study model. Full sized presentation models have also be used, for example in the design of the Washington D.C. subway system as documented by lighting consultant William Lam. [Lam, William. Perception and Lighting as Formgivers for Architecture. New York: McGraw-Hill, 87-93.] Existing computer graphics systems also are capable of making three-dimensional study models and to an increasing extent are capable of producing very realistic presentation images based on three-dimensional computer models. It is important to be clear on the difference between all of these and the more provisional and mutable sketch models that are the object of this discussion.

There are four criteria, adequately met by the designer's traditional methods of sketching, that any working design system must meet. The system must be:

A successful "sketch" interface for computer modeling will meet these criteria.


Design and Computing

An Interface Issue

Because the character of the interaction between a designer and his design media is crucial to the design process, functional computer-based sketching is largely a matter of human/computer interface. Recently developed computers with bit-mapped displays and graphically controlled windowing user environments present new options for interfaces to three dimensional modeling and rendering. In light of the new potential of these graphic workstations we can rethink the process of entering architectural information, and develop an interface that will be easy, fast, and fluid enough to allow real design sketching in three dimensions.

The complexity of the data to be entered and the need to allow design work to proceed in an intuitively oriented mind state present conflicting requirements for design software. The prevailing expectation seems to be that these factors will be accommodatYessios, Chris. "What Has Yet to be CAD." ACADIA Workshop Proceedings, October 1986, 29.ed by building artificial intelligence into architectural graphic systems. This would help by allowing the computer to develop a significant part of the building description data set. This in turn would allow the designer to keep his thinking at a high level, on the design rather than the designing. In general this seems to be a plausible approach, though certainly not an easy one. However, the decision programming for the computer is a formidable problem. More importantly, deterministic effects on design must be avoided if the tool is to be acceptable.

I believe we need to use a different approach, developing alternatives to artificial intelligence programs that will work by facilitating the designer's work rather than by supplementing it. This can be done by improving the interface between the designer and a modeling space displayed by the computer.

The design process can break down from a failure to generate alternatives. Failure of the design process is more often due to a faulty choice among alternatives, caused by a failure to visualize them realistically (honestly, in effective detail) or caused by some other failure of evaluation. In Design for the Real World Papanek, Victor. Design for the Real World. London: Granada, 1974, 142., for example, Victor Papanek states that architects, engineers, and draftsmen fail to draw a correct perspective view of a particular object, when given front and side elevations, at a higher rate than people without a measured drawing background. Apparently this perceptual breakdown is due to habitual rather than explicit projection in the designer's mind's eye. The incorrect acceptance of an existing alternative will also short circuit the generation of new possibilities.

Choosing among alternate design paths is too important and not well enough defined to delegate to the machine, but the machine could help substantially with the visualization if it could hold images to hand, and allow the right kinds of subtle adjustments to those images without getting in the way.

The complexity of editing building models as compared to text handling, numerical analysis, or even two-dimensional graphic design work remains a substantial barrier. Ability to manipulate text effectively with a computer is very recent, and it is dependent on millennia of language development for the simple abstractions used. Creative buildings remain with one foot in the realm of sculptural art, where the issues belie abstraction. A building exists as a full bodied entity out in reality. Good design solutions are even likely to be anti-abstraction, and we should reject any tool that reduces our ability to work with the fine-grain character of a project.

The best defined set of existing abstractions in architectural representation is the conventions of working drawings (figure 6). However, construction documents describe a set of rules and guidelines for the building process rather than the experience of the building and space. That full sense of experience is what needs to be held to hand by the media of design.

Figure 6. These standard symbols are part of the abstract language of working drawings.

That is exactly where existing applications fall down. The cumbersome modeling process gets in the way of the design process, and the toolkits don't readily support the sorts of model adjustments an architect is likely to want. More specifically, the standard CAD applications fall down on each of the key issues distractiveness, appropriate specificity, malleability, and subtlety. In the traditional sketch these are handled by the directness and expressiveness of pencil on paper. We must alter the methods of modeling in graphics software to achieve that level of directness and nuance. We must provide the same fundamental attributes using physically different but conceptually parallel methods.

With an appropriate interface, solid sketching could become a new, powerful kind of design tool.

Existing Two-Dimensional Interfaces

Most of the research on human/computer interaction has focused on two-dimensional modes, which represented the frontier of development in such interactions until quite recently. However, a sufficient body of work now exists to allow the generalized definition of advanced approaches to both text handling and graphic manipulation in two dimensions. Specifically, classic studies using several variations of the keystroke model for interface analysis demonstrate the effectiveness of mouse-based graphic interfaces for a variety of fundamental selection and positioning tasks. For example, graphically oriented text editors successfully reduce the number of movements required to perform even simple manipulations of text, as compared to typed-command driven interfaces. [Foley, James D., Victor L Wallace, and Peggy Chan. "The Human Factors of Computer Graphics Interaction Techniques." IEEE CG&A, November 1984.] For complicated graphic tasks the superiority of a graphic interface is only greater.

A good tool is sufficiently general to do a variety of work and to support creative extension, yet specific enough to facilitate the accomplishment of particular tasks. For example, the carpenter's plane is a powerful tool for smoothing wood exactly because it is constrained from making deep cuts. Innovators in the hardware side of interface research have tended to be seduced by the apparent power of very general tools. An extreme example of this, at least from a modeling interface point of view, is the sensor glove. [Foley, James D. "Interfaces for Advanced Computing," Scientific American, October 1987, 126-135.] These wired gloves provide the amazing facility of putting one's hand directly into the modeling space. Once there, however, the hand is no more useful for drawing than fingers are for painting. Exotic input hardware may be a distraction from the fundamental conceptual problems of three-dimensional manipulation. "I think we can build anything we can conceive of. So we're terribly knowledge-limited today." [Jerome Weisner, in Brand, Stewart. The Media Lab. NewYork: Viking, 1987, 154.]

Within the class of graphic input devices, the conventional, economical mouse has advantages for many tasks. Since the mouse is inherently a relative positioning device, it tends to focus attention on the graphics display. The window on screen contains all the location information, as opposed to sharing it with the location of a stylus on a tablet. This works synergistically in an interface that provides continuous on-screen feedback of selection and editing status. Tablet menus, in contrast, dependent on the absolute positioning capability of a digitizing tablet, divide user attention between the screen and a tablet map. In the days of slow vector displays, when on-screen text was a liability, having a menu on a tablet was an asset, but on a windowing workstation an absolute positioning device like a graphics tablet is optimal only for the generally minor function of hand digitizing.

One important aspect of the materiality of drawing a line on paper, and understanding how it fits in a drawing, is a sense of duration that translates to a feeling of length in the mind's eye. This implies that it is more meaningful to the designer to drag lines into a drawing, rather than just point to disembodied end points. The relative position information supplied by the mouse, without zeroing problems, is ideal for this approach to input. The sensate duration/length connection to object manipulations is equally relevant to the fundamental editing operations of resizing and repositioning, both of which can also be performed by dragging. Pointing combined with dragging works well for indicating a simple selection set.

The Apple Macintosh has made a commercial success of the graphic mode of interacting with a computer. In the Macintosh environment, normal hand and eye skills are used to simply accomplish graphic transformations that would be complex to describe in words--even complex in the special words of a tailored command language.

As indicated by personal use as well as commercial success, the mouse-based Macintosh interface is a good guide for how to handle two-dimensional graphic tasks, including manipulation of text and two-dimensional drawing. The Macintosh interface is a careful and creative implementation of many years of academic and industrial research in human factors and ergonomics, presented in an unusually consistent form on a relatively affordable platform. It does not provide a perfect solution to the problems of human-computer interaction. In particular, it has not been developed to do the key tasks of three-dimensional work. Nonetheless, the spirit of the Macintosh interface remains a guide into this uncharted territory. Both the method and the machine of the mouse can be effectively extended into three dimensions, providing an inexpensive yet appropriately flexible input and manipulation system created completely through thoughtful software design.

In the Macintosh interface in general, and in Macintosh MacDraw-type graphic applications in particular, operations are performed with noun-verb, or select-then-operate sequence. This is a natural arrangement for design work. First attention is directed at an object, and then and action is taken regarding the object. Command line interfaces tend to use a verb-noun, or command-argument syntax. The verb-noun order often puts the user in the position of remembering the object of the next action while simultaneously trying to recall the command to excute the action. This is represents a forced allocation of user attention for the convenience of the machine.

There is a difference in the fundamental program outline for these divergent interface approaches. The command-argument approach feeds input to program subroutines in a relatively simple, linear manner. In contrast, a select-then-operate application, especially if using a mouse for graphic input, is structured around a main loop waiting for input events. These programs must be able to respond to events in any sequence.

In Macintosh-style applications, graphic selections are made either by directly clicking on an object or by a selection range, defined by dragging diagonally across the desired area in the current view. Subsequent commands are applied to the current selection set. [Apple Computer, Inc. Human Interface Guidelines: The Apple Desktop Interface. Reading, MA: Addison Wesley, 1987.] Simple objects such as lines, rectangles, and circles are created with one mouse gesture, anchored at the location indicated by the mouse-down of a drag motion, at the size, direction and proportion indicated by the mouse-up concluding the gesture. A selected object may be resized by dragging one of its handles, and it may be moved by dragging it other than by a handle.

In addition to this basic pattern, control options are provided through a small set of modifier keys. Mouse clicks on objects with the shift key depressed invert the selection state of the hit objects--that is, where a new click would normally drop an existing selection set and start a new set, a shift-click will add non-selected objects to the existing set and delete already selected objects from the set. The shift key also modifies resize and move operations. Unmodified dragging of a rectangle handle will resize the object to a new proportion, keeping the four sides in a constant angular relationship (shape), and with the shift key depressed resizing is by handle dragging is constrained to the existing given proportion.

Figure 7. The MacDraw interface. Note object handles on selected rectangle.

[Foley, J. D. and Van Dam, Andries. Fundamentals of Interactive Computer Graphics. Reading, MA: Addison-Wesley, 1982.] [Knaster, Scott. Macitosh Programming Secrets. Reading, MA: Addison-Wesley, 1988, 21.]

Unmodified dragging of an object off a handle will move it in any direction, but depressing the shift key constrains the movement to orthogonal paths. Depressing the shift key during the creation of a rectangle constrains the proportions of new object, producing squares sized by the mouse gesture. The ellipse tool when constrained creates circles. A less standardized but similarly useful convention is that the option key allows handle dragging to reshape rather than resize objects. For instance, dragging on the handle of a rectangle with the option key depressed will move just one handle, creating an irregular four sided polygon. Together, these modifier conventions allow a very simple set of tools to provide a wide variety of specifically controlled drawing actions.

Equally important to the control patterns is the system of continuous feedback cues provided to the user. Selection is indicated by displaying the handles of the selected objects (figure 7). Selection ranges are shown during their definition as dotted line rubber-banding boxes. Feedback during object creation consists of continuous display of the size and location of the proto object, and resize and reshape operations are previewed similarly

The combination of noun-verb syntax, simple tools with modifiers, and visual feedback, supports a gesture-by-gesture way of working. This lets the user work on a problem in a chunk-by-chunk fashion, as he is accustomed to, with the computer responding in parallel. Verb-noun interfaces, interfaces requiring specific commands for each manipulation, and interfaces that demand the input of more abstract parameters all require a greater degree of premeditation for each adjustment to a drawing. This means a higher cognitive loading, and therefore less attention for design.

The basic requirements of the create, select, and edit operations don't really change when we jump to three dimensions, and the kinds of visual status feedback provided by the Macintosh drawing tools would be just as invaluable in a three-dimensional model space as they are on a two-dimensional screen. The material sense of duration is just as informative of vertical distance as it is of distance across a plane. We should provide for direct manipulation of objects and controls, with continuous graphic indication of current tool, position, and selections.

Existing Three-Dimensional Interfaces

Existing three-dimensional applications provide several methods for creating or defining objects in a model space, but a survey of a few representative applications will show that drawing in space remains a bottleneck in the modeling building and editing process.

AutoCAD holds a dominant share of the CAD market, and as the market leader it provides a standard for comparison. [AutoCAD, version 2.62.] The program has been gradually extended from its strictly two-dimensional origins to a usable surface-modeling application, with substantial rendering capabilities available in the AutoShade extension application. AutoCAD contributes the concept of "filters," which allow multiple mouse hits on surrounding objects to cumulatively indicate a target point point in space. Each hit is filtered according to a preceding menu selection or typed code so that only the one or two indicated axial values is added to the point being defined. Yet this makes a multi-step procedure of indicating even a single location in space. Constructing a three-dimensional model in AutoCAD is an awkward process, requiring careful strategizing to get a satisfactory construct into the under-powered surface model data structure, using a limited and non-architectural set of tools. The AutoCAD process is primarily directed at making wire-frame models and defining surfaces, which can be surface-rendered in AutoShade, a separate add-on program.

The McDonnell Douglas Graphic Design System (GDS) is an example of a relatively mature minicomputer CAD application with sophisticated modeling modules. [GDS, version 4.9.] These modules support user-definable multiple active views, constructive solid geometric and boolean operations on named solid objects, and surface rendering with shading and shadow casting. Within the modeling module, all editing functions are available in any view. However, limitations in the methods available for graphically specifying locations in space, together with an overly general set of modeling functions, preclude effective use of this powerful environment for conceptual building design. In the GDS three-dimensional space, as well as in other existing three-dimensional CAD environments, direct mouse hits are always interpreted as the point of intersection between a line of sight extended back from the cursor screen coordinates and a two dimensional "working plane." This means that a three-dimensional object cannot simply be drawn with the mouse--almost any action extending off the two-dimensional working plane requires keyboard specification of coordinates. Outlines are drawn in the working plane, and then extruded into solids with an additional multi-part command. When the designer has to stop and type x,y,z coordinates and displacement vectors, the "design mind" grinds to a halt. Entering numerical coordinates for the third dimension exemplifies getting too specific too early.

The GDS modeling toolkit includes a very powerful set of boolean operations, allowing the development of very complex forms by spatial unions, intersections and subtractions. From the abstract point of view of constructive solid geometry, these functions, together with move, rotate, extrude, and pocket, make a very complete set. However, certain basic manipulations of interactive architectural design modeling, such as changing the pitch of a roof slab, are very difficult to achieve with the current GDS tools. The mathematical functions of solid geometry are implemented in general with an emphasis on creation of forms rather than adjustment of form, to the detriment of model editability. The sophisticated but relatively static models that result do not support the evolution of design concepts. They are valuable for understanding existing concepts, but they may even impede the development of subtle design improvements.

Among three-dimensional applications for the Macintosh, Schema in particular has been presented as a sketch modeling system, and it implements several advances. [[Van Norman, Mark. "A Digital Modelshop: The Role of Metaphor in a CADD User Interface," Design Computing, (1986) Vol. 1, 95-122] The key function of creating a cuboid illustrates a creative approach to defining a three-dimensional object on a flat screen. Drawing a two-dimensional polygon defines the plan of a block. When the polygon is closed, by a double-click on its starting point, a rubber-banding line appears, and the length of that line as defined by one more mouse hit indicates the vertical dimension of the new object. This is a simpler, more graphically oriented implementation of the classic extrude function found in GDS. Since the ending segment indicating extrusion length is rubber-banded in the working plane, in a meaningless direction, the Schema version still does not allow the direct visually-determined sizing of new objects. Noticeable cognitive dissonance occurs from using a vector to specify only a length, although this decreases with familiarity. A greater limitation of the Schema environment is a lack of functions for editing blocks after their creation. As in GDS, this lack stunts the potential of the environment for design. Work is done in plan rather than in axonometric or perspective views. In addition, the careful applied model shop metaphor in Schema clashes with the intuitive quality of early design. The reference to relatively deliberate tools such as mills and bandsaw is more appropriate to making study models, as one would be likely to do in a physical workshop. These tools do not provide helpful metaphors for the creation of conceptual models because even in the real world their use requires a high degree of premeditation.

There is a further difficulty in defining a three dimensional sketch interface by metaphor. Since little design is currently done in three dimensions, the design tool metaphors sought by the creators of Schema are just not available. First principles of design method derived in a two dimensional context are the main available source of truly design oriented tools, for better or worse. Attempting to find direct three-dimensional metaphors for design tools, the creators of Schema have found three-dimensional metaphors that provide little in the way of design-orientation. [Mitchell, William. "Solid Modeling and Volumetric Composition in Architecture," Design Computing, Volume 1, 1986, 123-135.] They have succeed in making a non-numerical modeling interface, but it remains abstract and relatively inflexible.

MacArchitrion, a relatively new CAD application for the Macintosh, achieves a relatively high level of sketchability. [Gimeor, Inc. MacArchitrion: Version Series 3.5E through 3.7E application and manuals, Washington D.C. 1988.] In MacArchitrion, cuboid blocks are created casually, drawing them in plan with base, height, width, and justification controlled by settings in an editable dialog window (figure 12). Once created, cuboids may be moved and edited quickly into desired forms using an architecturally relevant tool set. For example, one mouse hit near the edge of a block will incline one or both horizontal surfaces to a specified slope, or increment the pitch of the surfaces by a given amount. This provides a very easy way to make and adjust roof surfaces and slabs. The variety of simple editing functions such as incline, including cut block, align faces vertical or horizontal, and change base, height, width, length, or any corner location of a block, makes for very malleable models. Linear and radial array functions support easy creation of repetitive elements, repeating both blocks and openings in blocks with independent x, y, base, and height increments, and with automatic mitering of blocks in radial or spiral arrays. The powerful MacArchitrion environment falls quite a bit short of ideal, however, because manipulations are only possible in the plan view. Although an axonometric or perspective view is always only one keystoke away, the editing manipulations themselves must be performed "blind" of the third dimension. Overlapping block ambiguities are resolved by a "block order" setting, which can cause some surprises. Even the planned upgrade to allow working in other orthogonal views will not provide the combination of direct manipulation with continuous feedback that a free design sketching environment requires.

Unfortunately, no three-dimensional application I have been able to examine provides an environment comparable to that of the original MacDraw, typified by the absence of a "move" command. "Move" in MacDraw is an inate function of the model space, accomplished by one mouse hit to select and another to drag. No command is required to select, move or resize an object in MacDraw (or other applications with a similar structure). This kind of functionality, with interface transcending "features," is central to creating an actual system for computer-aided design. This kind of functionality has not yet been provided in a three-dimensional modeling application.

The lessons of the two-dimensional environment have not been extended to three-dimensional interfaces. Indirectness of manipulation has been enforced by the use of two-dimensional input and two-dimensional display for three-dimensional modeling, leaving a barrier between the designer and the model. I will describe a series of interface constructs designed specifically to address the special character of spatial modeling, circumventing this basic obstacle and allowing direct manipulation of solid models. The goal is to be able to make and adjust solids in model space as freely as hands can adjust modeling clay or cardboard, yet with greater subtlety, with record, recall and revise capabilities, and as a consistent step in a continuous process of model development. This will allow real architectural design in three dimensions.

Three-Dimensional Sketching

The survey of some existing applications suggests that careful consideration of the qualities of traditional design tools is needed for the delineation of a new modeling environment. Specific interface approaches that will help make the process of orientation and data entry in three dimensions intuitive rather than overwhelming need to be defined in terms of fundamental principles.

The fundamental need is to draw easily, whether in space or on a single plane. The most basic aspect of drawing is marking points on a page. In order to apply creative design work to the difficult problems, we need tools that are simple. We need to be able to mark points in space with an intuitive hand-and-eye ease similar to marking paper with pen or marking a screen location with a mouse. Once this most basic and essential facility is provided we can develop a structure of objects and methods that will allow for modulation of design ambiguity despite the design's underlying basis in a precise digital model.

Even the difficult to learn, drafting oriented, command-line driven CAD applications allow easy marking of points on an x-y plane using a pointing device. However, an analogous mechanism for accessing points in space has not been provided, even in recently introduced applications with two-dimensionally sophisticated graphical interfaces. Due to the restrictions of existing interfaces, the basic act of indicating points has remained a barrier to intuitive modeling. With current tools, the third coordinate must be entered from the keyboard as a numerical value. A numerical dimension is unnecessarily precise for early design work, and conjuring up a specific value immediately puts the designer off track. Instead, the designer should be able to place blocks in space manually, sizing and positioning them by eye. Visual focus should stay on the model itself, without shifting around to accommodate control arrangements.

The key proposal for a solid sketching environment is mouse control of three dimensions instead of just two.

The mouse is basically a two-dimensional positioner, and has been used as such in modeling applications. But it is apt for direct manipulation interfaces, and it can be augmented to allow direct manipulation in three dimensions. The sketch system proposed here uses mouse movement on the table to indicate horizontal cursor movements, leaving x-y in-plane movements as they are in typical two-dimensional systems. This will allow maximum transfer of experience and coordination skills from two-dimensional graphic work. Control of vertical cursor movements is added by means of z-axis elevator buttons, using a pair of letter keys for up and down cursor elevators, or using the second and third buttons on a three button mouse. Upward displacement is indicated by left-button-down-duration and downward displacement by right-button-down-duration.

By lifting the cursor free of any particular "working plane," the direct control of three-dimensional coordinates with the mouse will result in substantial confusion if improved orientation tools are not provided simultaneously. Keeping track of cursor location in three dimensions can be distractingly difficult in itself. Therefore, an adjunct piece of the interface would be a three axis crosshair. Even the spatial position of a full-box cursor can be ambiguous, so to highlight the location of the hot point of the crosshair, additional cues should be available. A set of background grids (a floor grid for the x,y plane, side wall grid for the y,z plane, and a back wall grid for the and a back wall grid for the x,z plane), with the points of intersection between the crosshair rays and each background grid highlighted, would aid in identifying the cursor location (figure 8). In addition to graphically indicating cursor coordinates, background grids could clarify the three-dimensional proportions of relatively orthogonal models, although they would be less useful with softer forms.

Figure 8. Sketching environment, showing three-dimensional crosshair with background grids, in perspective projection.

In some cases background grids may be objectionable, or ineffective, so consideration should be given to more technically demanding approaches to position indicating, such as depth cueing by relative intensity or indication of crosshair/object overlaps. The latter would require some degree of real-time hidden-line calculation, but with reversal of crosshair color to indicate "insideness" or "behindness," it could be very helpful. In addition, given ultra-fast intersection calculation, the point of crosshair penetration of object surfaces could be highlighted. The user could be given some control over the speed/alignment tradeoff by a variable setting for crosshair length--shorter means fewer objects to compare against, and longer means easier alignment with distant objects.

Some shape concepts are most naturally captured in outline, as simple planar wireframes or as sectional loops for extrusion. Therefore, it should also be easy to draw two-dimensional graphics in the model space in various orientations. To support this, the mouse table-top movement can be realigned to another plane in the model space--a temporary working "desktop" in section and elevation as well as in plan.

Figure 9. The GDS PLANE command provides a powerful but complex means for readjusting three-dimensional axes.

For instance, a cross-section of a complex extrusion perimeter can be drawn directly in the location and orientation in which it will be used, without the need to construct snap objects, and independent of the current viewpoint. With existing toolkits, drawing out of the horizontal plane requires working plane rotations which are disorienting and tedious to execute (figure 9). Due to the fundamental lack of a three-dimensional cursor, new elements can only be drawn in views orthogonal to their plane of creation, or by snapping to pre-existing objects. Existing approaches require a one-to-one relationship between the working plane and the two dimensional cursor plane. In the three-dimensional sketching environment, a single icon with multiple facets can allow a single click to orient mouse x-y to any of the three major planes for more facile drawing. A direct method for orienting the working axes non-orthogonally to the world axes would be a valuable extension to this, and three-dimensional cursor control opens many possibilities. For example, the realignment of axes could be handled by dragging the primary crosshair in the model space with a temporary secondary crosshair.

Direct x,y,z cursor control, a three-dimensional crosshair option, and iconic selection of working plane constitute the fundamental tool set. These establish an extensible paradigm for direct manipulation of solids in space. The combination of direct movement in space with clear orienters in space does the main job of liberating the architect model maker from the distraction of numerical coordinates and computer-oriented manipulations in the sketching environment. Concern is shifted from the keyboard to the modeling space and from the dimensions of forms to the forms themselves.

In addition to these fundamental tools, methods are critical, and objects need to be invented in connection with the fundamental editing method. As demonstrated for two-dimensional work by MacDraw, it should be possible to accomplish the basic editing operations without having to enter commands. For example, an object may be resized by selecting an object handle and then dragging it in any direction with the three-dimensional mouse. The select-then-operate syntax is as appropriate to three-dimensional modeling as it is to MacDraw. Object selection is indicated by handle highlighting.

Figure 10. Selected rubber block with three-dimensional cursor and highlighted handles.

Similar to a "rubber banding" line in a two-dimensional environment, objects of this method may be referred to as "rubber blocks" (figure 10). Although not all objects will be cuboid, the handles of an object may be conveniently displayed on a cuboid pattern representing the state of the object's coordinate "box."

In the context of the three-dimensional cursor, the handles for adjusting rubber solids can be very similar to the handles of rubber polygons. A polygon may be considered a flat block, such that grabbing a corner handle and dragging it in the x-y plane resizes the polygon, and dragging that handle in the z direction (using the elevator buttons) will extrude the polygon. Stated differently, dragging in the z direction will resize the block from zero to some finite thickness. Clicking on an edge-midpoint handle allows edge movement for object resizing, and edge movement constrained by the shift key provides independent movement of faces. Option-dragging an edge mid-point handle allows simple reshaping of the object, and shift-option-dragging porvides controlled reshaping based on orthogonal edge displacements.(figure 11).

To provide the user with status feedback, each time a handle is grabbed the handle itself should change from solid fill to bold outline, and the lines in the wireframe which become adjustable should change to a dotted representation. Clicking or boxing an unselected object selects it, as indicated by the appearance of its handles. Mouse-down on a selected object other than on a handle grabs it for relocation, and the next mouse-up places the object.

In addition to the resizing and reshaping operations accomplished by dragging handles of a selected object, a further range of size and shape adjustments can be accomplished by selecting for handles rather than objects. The option key may be used to modify mouse events during selection (by either clicks or drag-boxes) so that only the indicated handles themselves are selected. For general architectural purposes, for example, openings in objects should be mapped isotropically onto the object coordinate system (i.e. when a wall is made higher, the openings stay the same). But in this environment, some of the stretch operations not supported well by isotropic mapping of openings (i.e. when a wall is lengthened, the windows elongate with it) can still be supported in a powerful and transparent manner by means of bulk operations on the selected handles from objects and openings, as opposed to operations on the objects themselves. The distinction allows the modeler to receive both modes of behavior from a single class of objects, without having to define any additional parameters, because the transformation mode is defined by the mode of selection.

The combination of the tools and the methods determines how modeling tasks will be accomplished. It is notable that scaling and translation, two of the three canonical object transformations, are accomplished without the need to give any command. This provides a

Figure 11. Resizing and reshapeing operations on rubber blocks.

profound response to the menus-vs-icons-vs-keyboard controversy. A typical way to create a cuboid would be to move the cursor to the desired starting elevation, if necessary, then drag out the diagonal of the block in plan, drag the outline upward with the up button, and release the button at the desired block height. Continuous display of the changing shape of objects during transformations is critical. For objects or groups too complex to redraw interactively,it might be adequate to continuously display the object or group coordinate box,during transformations. It is less critical whether the associated syntax follows the original MacDraw button-down/drag/button-up pattern or the click-to-start/move/click-to-finish alternative. The down-and-drag approach gives the most emphatic space-durational feedback, but the click-click syntax lets snap codes be reset in mid-transformation with maximum ease. The first favors freehand sketching, and the second favors precision model editing. The best interface option would be to allow user switching between the two methods.

Most CAD applications ignore open space (non-object) hits, but event-driven, Macintosh-style applications interpret open hits as the start of a selection-box-define procedure. Thus a blank hit undoes the current selection set. The three-dimensional cursor allows the definition of a three-dimensional (space-based) selection box, with improved selection resolution compared to two-dimensional (view-based) selection boxes.

It is not always desirable to specify every dimensional parameter graphically for every block. Tools should allow quick and independent fixing, resetting, and releasing of absolute or relative values for most creation and modification parameters. For example, a cuboid with base set at z=0, height set at relative 8'-0", width set at relative 6", and floating length, can be used like a line-type for quick drawing of ground level perimeters and partions. MacArchitrion implements a powerful but somewhat fussy approximation of this, using modal dialog boxes extensively for setting parameter groups (figure 12).

Figure 12. MacArchitrion dialog box for setting block-creation parameters.

The three-dimensional sketching environment supports multiple constructional approaches to basic architectural modeling tasks. One could, for another example, start in the xz plane, draw a wall outline in section, extrude the wall by clicking on a handle and holding down the left button for a moment, then click again to end the extrusion, hit an icon to select yz as the working plane, drag diagonally to define a window outline, hit a "remove" icon, and (with an orthogonal constraint, put into effect by holding down the shift key) drag the window outline perpendicular to its plane to cut a pocket through the wall. Alternately, a mouse hit following a hit on an opening icon could place an opening with its own set of handles in the middle of the wall. This could then be dragged to various positions in the wall, or resized by dragging its handles, just like a block in space. (The opening should be copyable, as well.)

More complex or specialized transformations may be more easily applied by command, given by menu selection, icon hit, or keyboard entry in the same noun-verb pattern. Useful tools for this environment along those lines include "create walls," which will convert the solid blocks of a massing model into hollow blocks with specified wall, floor and roof thicknesses; "create opening," which simplifies the placement of doors and windows by cutting an opening, with a given height and width, through a block at a location indicated by a single mouse hit; "rubber windows," which allow opening adjustment by three-dimensional dragging of handles on the frame of the opening; graphic selection of parametric primitives from a library, for insertion at indicated locations with interactive parameter editing using dialog boxes or handle dragging; and incline selected surface(s).

Other commands would be useful for refining rough models. As project design progresses from conceptual to developed, the ambiguities that were originally helpful need to be progressively worked out, and a few particular tools would greatly facilitate that refinement. These tools could make possible the seamless transition out of schematic design into design development. One example of this kind of function is "make-square"-keeping-indicated-point-fixed. Addtional tools for cleaning up sketch models could include "align to indicated" surface and "align to average" of selected surfaces. A "scale space to model" command (including snap grid setup) will ease the initial transition from a dimensionless conceptual model to a developing measured model.

The actual programming to implement three-dimensional cursor control should not be difficult, consisting primarily of a few new input handling routines. However, it could be a considerable task to develop model and object data structures to support resizing and reshaping by handle dragging. A possible approach would be to treat each object as an unchanging fundamental primitive, which gets expressed parametrically in terms of the distortion of its axes. Each block in a model would consist of an instance of an original block, together with the transformed version of its own coordinate system and the location of its object coordinate origin in terms of the world coordinate system. Moving the block would reset its location in the world, and moving a handle would reset the object's expression parameters, or object coordinates.

This structure provides a basis for supporting named types of objects with actual sizes and shapes varying among instances of a type. This in turn lays the groundwork for increasing the analytical information content of the model. If the types are readily invented and readily selected from a graphically accessed library (i.e. of various walls, roofs, doors, etc., perhaps with attributes already attached) this extra information can go into the model with very little additional cognitive load on the modeler.

Graphic controls for viewing parameters are also important. For example, a scrollbar type in-out view box positioner should be added to the x and y scrollbars commonly provided on graphics windows, as well as scroll controls for rotating and wheeling eye point and view point. Since the existing x and y scrollbars refer to x and y of the view-frame (window) rather than x and y of the model space, it may be necessary to design an entire additional set of three-dimensional scrolling controls. The two-dimensional zoom-to-window command can be replaced with zoom-to-box, as dragged out by the three-dimensional cursor. If redraw rates are sufficiently fast a continuously variable zoom valuator would be useful.

The interface described meets the first three criteria described in an earlier section. It provides a modeling environment which is relatively undistracting, due to a minimum of commands for a maximum of manipulations; which is not overly specific, due to fully graphical direct manipulation of objects; and which creates highly malleable constructions, capable of interactive adjustment.

The fourth criterion, relating to the expressive subtlety of the medium, is less tangible, but equally critical to the ultimate design utility of the system. The ability of the solid sketching environment to hold nuances of design meaning comes from its natural adjustability. Where the designer zeros in on the form in his mind's eye using paper by drawing over and over it, the solid sketcher can zero in by a little push here and a little stretch there. This difference between paper and computer sketching is somewhat similar to the difference between cardboard and modeling clay sketching. A cardboard sketch has the richness of an ambiguous roughness, but it is not so delicately adjustable, while clay yields to very slight adjustments. The solid sketching space adds save, recall, and compare to that subtle adjustability.

The simple but powerful combination of a three-dimensional cursor, position indicating bounding planes, direct hits in three dimensions with mouse control, and rubber block solids, will provide the designer with a unique plastic three-dimensional sketching space with the flexibility of clay, the easy planes of cardboard, and the immaterial freedom of paper-based sketching.


Putting It All Together

Looking Forward

Current elements provide the basis for a radically new possibility--accessible tools for the quick development of a true three dimensional model of a designer's mental images. The creation of this model in the electronic database of the workstation brings other possibilities with it.

Command procedures can be written to aid the refinement of a sketch model into a clean study model. Seamless progression of design development in three dimensions will become more possible with the aid of parametric object substitutions. Design development can proceed quickly and smoothly toward approximate working-drawing detail when building systems are included according to their governing equations. (Notice nonetheless how these facilitate design development toward detailed drawings, rather than conceptual design.) Builders can look forward to doing highly accurate estimates with a lot less drudgery as non-graphic data get built more smoothly into the overall project model.

Easier three-dimensional modeling will help take the design further, to a greater level of detail, before dropping down to two-dimensions for annotation. This could cause a shift in the project staffing curve. If designs are clearly defined by comprehensive three-dimensional models, the working drawings can be started at a later stage, and they may proceed more smoothly. Both effects will tend to shift effort out of production into design and design development. This would be a more welcome benefit from computerization than just faster drafting.

At the point when the next degree of detail would make the three-dimensional model of the project at hand too complex to work with effectively, two-dimensional cross-section cuts can be taken from the model. Thus the detailed three-dimensional model can be used as the source for the more specific drawings needed to meet the still contrasting needs of presentation graphics and construction documents. Each stage of design development can build upon the graphics of earlier stages, refining and extending its findings.

As a project moves into construction document production, an orthogonal view of the model can be saved as linework. Detailing of the cut images can be continued in two dimensions, using conventional "flat CAD," with no need for massive reentry of information. The resulting black and white graphic may be elaborated with notes, symbols, and dimensions, and may be reproduced easily. This development of a drawing in turn parallels the detailing of a design from the architects' refined conception into a carefully specified set of instructions.

Branching in the direction of presentation, a computed color rendering of a model can be added into a scene previously photographed, digitized, and loaded into a paint program where further adjustments can be made. Combining a synthetic building image with an actual context image parallels the basic projective process of visualizing how a new building will fit in an existing site. Sketch modeling on the computer will be part of a general convergence of electronic graphics. Video tapes made easily during a site visit can be loaded as background into the model space, providing real context with unparalleled ease.

Another direction for graphics for architecture to extend is beyond the "simple" solid model to four-dimensional modeling, with real-time movement through design spaces. This kind of interactive motion, available now on expensive super-computing workstations, has the potential to revolutionize architectural presentation in the near future. High speed rendering could be used to simulate a time lapse view of sunlight changing over the course of a day or season, as well as walk or fly throughs. The realism and information density of design video could have a very important impact, as pre-construction visualization becomes an expectation of clients who have up to now been kept at some distance by a relative lack of experience at reading drawings.

Interaction about design is central to architectural education as well, and communication of spatial concepts can be difficult. The ability of a sketch modeling environment to illustrate options in three dimensions could make it a valuable teaching tool, clarifying teacher and student conversations about design. Students and teachers looking at orthogonal views of a project, whether drawn by hand or with computer aid, may not imagine the same three-dimensional reality. By giving the individual student a better sense of the actual spatial effects of his own design moves, sketch modeling can significantly enhance a crucial aspect of design awareness.


New approaches to modeling issues based on fundamental principles will retain their validity independent of new technology such as interactive three-dimensional output devices. Three-dimensional sketching is an approach to computer-based solid modeling derived from the fundamental requirements for a design environment, as inferred and extrapolated from traditional design aids such as clay modeling or pen and pencil sketching. It promises to combine the potential for fluency of the traditional design environment with a directness for recording and adjusting three-dimensional concepts made possible by the immaterial nature of a mathematical data base and by highly interactive projection and manipulation techniques.

Architectural design is a special activity. It relies on the intuitive, sensate, artistic sense of designers for the creation, visualization and evaluation of three-dimensional forms. The long-standing paper-based methods for generating, recording, and communicating the images of design support the process in several crucial ways, but they leave a fundamental dimensional gap between representation and reality.

Existing computer-aided design software provides for a three-dimensional design medium, but it fails where tradtional sketching is strongest. Relatively awkward model-making methods distract the designer from his artistic sense and force him to remain on an abstract, quantiative plane. If computers are to provide a useful design medium, the human machine interface for three-dimensional modeling needs to be streamlined.

Recent developments in user-friendly interface design suggest the cogent potential of powerful yet approachable direct-manipulation graphical interfaces. The simple mouse-controlled three-dimensional cursor provides the basis for extending direct-manipulation into three dimensions. With related innovations in modeling objects and modeling methods, it is possible to create a new kind of malleable, artistically potent computer model, in an accessible and versitile modeling environment.

This sketch-modeling environment, supporting both the artistic directness of paper-based sketching and the full architectural embodiment provided by three-dimensional modeling, will allow a subtly new approach to architectural design. The computer can really be a tool for conceptual aesthetic design, when with it we can draw in space.


Alexander, Christopher. Notes on the Synthesis of Form. Cambridge, MA: Harvard University Press, 1964.

Alexander, Christopher. The Nature of Order. manuscript, Berkeley, CA., 1986.

Apple Computer, Inc. Human Interface Guidelines: The Apple Desktop Interface. Reading, MA: Addison Wesley, 1987.

Attoe, Wayne. Architecture and Critical Imagination. Chichester: John Wiley and Sons, 1978

Bier, Eric A. Snap-Dragging: Interactive Geometric Design In Two and Three Dimensions . PhD thesis, University of California, Berkeley, March 1988.

Bolt, Richard A. The Human Interface . Belmont: Lifetime Learning Publications, 1984.

Bonta, Juan Pablo. Architecture and Its Interpretation. New York: Rizzoli, 1979

Brand, Stewart. The Media Lab . New York: Viking Penguin, Inc., 1987.

Buxton, William. "An Informal Study of Selection-Positioning Tasks." Proceedings of Graphics Interface '82 , 1982.

Greenberg, Donald. "The Coming Breakthrough of Computers as a True Design Tool." Architectural Record , September, 1984.

Fleig, Karl, Editor. Alvar Aalto . Zurich: Verlag, 1971.

Foley, James D. "Interfaces for Advanced Computing." Scientific American , October 1987.

Foley, James D., V.L. Wallace, and P. Chan. "The Human Factors of Computer Graphics Interaction Techniques," IEEE CG&A , 4(11), 1984.

Foley, J. D. and Van Dam, Andries. Fundamentals of Interactive Computer Graphics . Reading, MA: Addison-Wesley, 1982.

Hanks, Kurt, Larry Belliston, and Dave Edwards. Design Yourself. Los Altos, CA: William Kaufman, Inc. 1977.

Heidegger, Martin (translated by John Macquarrie and Edward Robinson). Being and Time. New York: Harper and Row, 1962.

Knaster, Scott. Macintosh Programming Secrets . Menlo Park: Addison Wesley, 1987.

Lam, William. Perception and Lighting as Formgivers for Architecture . New York: McGraw-Hill.

Mishkoff, Henry C. Understanding Artificial Intelligence. 1985.

Mitchell, William J. Computer-Aided Architectural Design. New York: Van Nostrand Reinhold, 1977.

Mitchell, William. "What Was Computer Aided Design?" Progressive Architecture , May 1984.

Mitchell, William. "CAD's Role in the Building Design and Construction Process." CAD/CAM Handbook , 1985.

Mitchell, William. "Solid Modeling and Volumetric Composition in Architecture." Design Computing, January, 1986.

Papanek, Victor. Design for the Real World. London: Granada, 1974 (142).

Perkins, David (ed.). English Romantic Writers . New York: Harcourt Brace Javanovich, Inc., 1967.

Pfortmiller, Larry. "Data Structures in CAD Software." Byte , June 1987.

Reisner, P. "Formal Grammar and Human Factors Design of an Interactive Graphics System." IEEE Transactions on Software Engineering , 7(2), 1981.

Rittel, Horst W. J. "Some Principles for The Design of an Educational System for Design." DMG Newsletter, Vol. 4, Number 12, December 1970

Schilling, Terrance G. and Patricia M. Schilling. Intelligent Drawings . New York: McGraw Hill, 1987.

Schon, Donald A. "The Architectural Studio as an Exemplar of Education for Reflection in Action." Journal of Architectural Education, 38/1, Fall 1984,.

Smarte, Gene and Nicholas M. Baran. "Face to Face." BYTE , September 1988.

Stitt, Fred A. Systems Drafting. New York: McGraw Hill, 1980.

Van Norman, Mark. "A Digital Modelshop: The Role of Metaphor in a CADD User Interface" Design Computing , (1986) Vol. 1.

Williams, Greg. "The Macintosh Computer." BYTE, February 1984.

Yessios, Chris. "What Has Yet to be CAD." ACADIA Workshop Proceedings , October 1986.

Kevin Matthews | Architecture Computing | Architecture Foyer | Index
This document is provided for on-line viewing only, except as printed by Author.
© 1988-96 Kevin Matthews, All Rights Reserved.

http://faculty/matthews.kevin/mp3.02b.html - Posted KMM 96.02.05, rev. 96.02.11